Longitudinal Studies of Sleep Disturbances in Parkinson’s Disease

Purpose of Review Sleep disorders are among the most common non-motor symptoms in Parkinson’s disease (PD). Recent longitudinal studies of sleep in PD have utilized validated sleep questionnaires and video-polysomnography performed over multiple time points. This review summarizes existing longitudinal studies focusing on the prevalence, associations, and changes of sleep disorders in PD over time, as well as the methodologies used in these studies. Recent Findings Fifty-three longitudinal studies of sleep in PD were identified: excessive daytime sleepiness, insomnia, obstructive sleep apnea, rapid eye movement sleep behavior disorder (RBD), restless legs syndrome, and shift work disorder were studied in addition to other studies that had focused on either multiple sleep disorders or broadly on sleep disorders as a whole. The prevalence of sleep disorders increases over time and are associated particularly with non-motor features of disease. RBD is now considered an established prodromal feature of PD, but other sleep disorders do not clearly increase risk of subsequent PD. Further work is necessary to determine if treatment of sleep disorders in PD alters disease symptom and their progression or reduces PD risk. Summary Longitudinal studies of sleep in PD have demonstrated a high prevalence of sleep disorders that are associated with non-motor features of PD which can increase over time. More work is necessary to determine if treatment of sleep disorders can alter the course of PD.


Introduction
Sleep disturbances are one of the most common non-motor symptoms in Parkinson's disease (PD). Although sleep disturbances are typically a feature of advanced disease, they are also recognized to be present both at the prodromal and early stages of disease [1]. A broad range of sleep disorders are seen in PD including rapid eye movement (REM), sleep behavior disorder (RBD), excessive daytime sleepiness (EDS), insomnia, restless legs syndrome (RLS), and sleep disordered breathing. Cross-sectional studies studying a range of sleep disorders in PD report a high prevalence of any sleep disorders that exceed 95% [2] [3]. RBD has been of particular research interest as the disorder is now recognized as a prodromal feature of PD and other Parkinsonisms and can precede the development of these conditions by up to 20 years.
In 2011, Zoccolella and colleagues analyzed 17 longitudinal studies and found that the prevalence of sleep disorders increases with disease duration, and that its presence could be indicative of a more aggressive disease course or associated with particular phenotypes within PD [4]. Since then, there has been a growing interest in the study of sleep in PD: recommendations for PD sleep questionnaires have been published [5], further longitudinal PD cohorts studied, and This article is part of the Topical Collection on Sleep. gold standard video polysomnography (vPSG) utilized in longitudinal studies to better characterize sleep. The purpose of this review is to critically review existing longitudinal studies reporting on prevalence, associations, and changes over time of sleep disorders in PD. We have also examined the relationship between specific sleep disorders and future risk of PD. Finally, we have identified limitations in existing methodologies. We have identified 53 longitudinal studies of sleep in PD from 1996 which are summarized in Table 1.
Generally, sleep disturbances in PD patients have been studied in isolation, in the context of a broad range of nonmotor symptoms, in cross-sectional or in short-longitudinal cohorts. Only very few studies have examined the frequency and coexistence of multiple sleep disorders in individual patients, and their longitudinal changes over time. Therefore, in the sections below, we have first described the longitudinal studies performed on individual sleep disturbances. We have then described the studies that have assessed the coexistence of multiple sleep problems in individual patients. Finally, we have reported some studies that have shown how vPSG features could be predictive of development of dementia in PD patients.

Excessive Daytime Sleepiness
EDS is defined as "the inability to stay awake and alert during the major waking episodes of the day, resulting in periods of irrepressible need for sleep or unintended lapses into drowsiness or sleep" [57]. In PD, most studies define EDS using the well-established, easily performed subjective Epworth Sleepiness Scale (ESS) questionnaire [58], rather than the objective mean sleep latency test (MSLT) [59]. The MSLT provides objective measures of sleepiness on a single day, but does not reflect a subject's average experience of sleepiness across a range of daily activities which the ESS attempts to quantify [60]. However, EDS is a symptom, with many possible causes including obstructive sleep apnea (OSA) which requires a respiratory sleep study or videopolysomnography (vPSG) for diagnosis. This is infrequently performed in longitudinal studies studying EDS in PD.
An epidemiological study from the Honolulu-Asia Aging study studying elderly male subjects showed that EDS was associated with an almost threefold increase in the risk of PD compared to subjects without EDS [12]. Other sleep features studied included insomnia, daytime napping, early morning grogginess, and frequent nocturnal awakening that were not associated with increased PD risk.
Six longitudinal studies have been performed investigating EDS in PD. In an earlier prospective cohort studied over a year, dopamine agonists (DA) were particularly implicated in causing sedation [13]. However, this finding has not been consistent across all studies with one study showing that DA did not alter EDS [10].
Subsequent longitudinal studies have focused on the prevalence of EDS in PD and its change over time: EDS was shown to be present at baseline [8,9], with most studies reporting increased prevalence of EDS [8,9,11] or increase in ESS scores over time [6]. Only one small study of 30 moderate stage PD subjects showed that ESS scores remained stable over 10 years [7] and also highlighted variability of the ESS scores. Other larger studies, however, have shown that EDS is a persistent and progressive feature in most patients [11].
Across the various studies, EDS was typically associated with non-motor features such as poor sleep quality, fatigue, anxiety, depression or axial, postural, or gait disturbances rather than motor severity [7,10,61], or cognitive dysfunction [61]. Other studies have shown an association between EDS with greater motor severity and male gender [10,11].

Insomnia
"Insomnia Disorder" is defined as the persistent difficulty with sleep initiation, sleep maintenance, or quality in an individual with adequate opportunity for sleep, resulting in daytime impact. It must be present for > 3 months and occur most days of the week [57]. Hence, the diagnosis is dependent on the subjective self-reporting of symptoms and the patient's perspective of impact on daytime function. Existing longitudinal studies of insomnia in PD have utilized different methodologies to ascertain the presence of insomnia, ranging from semi-structured interviews to standardized questionnaires, and have reported conflicting results.
The overall prevalence of insomnia in two longitudinal cohorts was not shown to change over time [16,17] but another study showed a slight increase in the prevalence of insomnia in a cohort studied over 5 years [15]. Significant fluctuation in the reported symptoms of insomnia over time was observed in a PD cohort studied over 8 years [17]: with the proportion of subjects having difficulties with sleep initiation or frequent awakenings increasing over time. Another group did not show an increase in prevalence of insomnia over time, but showed changes in the prevalence of insomnia subtypes: with the prevalence of solitary problems in sleep maintenance increasing in the first year but the prevalence of solitary sleep initiation problems decreasing over 5 years [16].
Epidemiological studies in the general population have established that the presence of insomnia or poor sleep quality results in an increased lifetime risk of depression; with females more likely than men to report insomnia in every age group [62], this is unsurprising as the same associations are well described in the PD population. Studies of insomnia in PD have been consistent in the association of insomnia with mood disorders: one study showed that baseline anxiety and depression predicted the development of insomnia in early PD [14]. However, unlike findings from the general population, the presence of insomnia at baseline did not predict subsequent affective symptoms. This discrepancy could be due to the shorter length of follow-up in the PD population studies compared to epidemiological studies performed in the general population. Other earlier mentioned studies have shown the association of insomnia with PD disease duration [17], depressive symptoms [15,17], female gender [17], motor fluctuations [15], and higher LEDD use [15].

Obstructive Sleep Apnea
OSA is a common sleep disorder characterized by snoring and recurrent collapse of the airways during sleep leading to unrefreshing night sleep and daytime sleepiness [63]. It is diagnosed on respiratory polygraphy or vPSG by the presence of apneas and hypopneas during sleep [63]. Risk factors include increasing age, male gender, and obesity and the incidence is rising across the western world [63]. As OSA results in hypoxia, inflammation, and oxidative stress [64], its role as a prospective risk factor for the development of PD has been of recent interest and studied by four different groups using data obtained from the Taiwan Longitudinal Health Insurance Database [19][20][21][22]. All studies reported that the presence of OSA was associated with an increased risk of PD (OR 1.35 to 3.54), differing according to the specific subgroup studied. These studies utilizing the Taiwan Longitudinal Health Insurance Database identified subjects with OSA using the International Classification of Disease, 9th edition, Clinical Modification (ICD-9-CM) codes for OSA. Polysomnography data, in particular the apnea-hypopnea index (AHI), were not available.
Although very large study numbers were available over 3 to 11 years, there were low absolute numbers of incident PD. The hazard ratio for the risk of PD was slightly elevated, suggesting that OSA is not a particularly strong risk factor for the development of PD. Further subgroup analyses showed that OSA in combination with coronary artery disease [20], stroke [20], chronic kidney disease [20], and insomnia [22] could increase the risk of PD. However, conflicting data with regard to increased risk for PD according to gender was found: one study showed that the increased risk of PD was only observed in males with OSA [19], while a different study did not find this association but showed that in females with OSA, the HR for PD was elevated at 3.54 [21]. Finally, another study with data over 11 years showed that females subjects aged 50-69 years with OSA had the highest risk for PD [22]. However, these findings have not yet been replicated in other study populations.
The symptoms of OSA are effectively treated with continuous positive airway pressure (CPAP) therapy. However, the effects of CPAP treatment on cardiovascular events and cognitive functioning in the general population remain uncertain requiring further study with larger randomized cohorts [63]. In a recent prospective cohort of 67 PD subjects followed up for a year, CPAP therapy was shown to stabilize motor function as compared to other patient groups who did not have CPAP therapy or were not diagnosed with OSA [18]. However, this cohort was small, and replication in future studies will be necessary.

REM Sleep Behavior Disorder
It is now well established that "isolated" RBD is a prodromal stage before overt neurodegeneration. In a landmark, multicentre study that recruited 1280 cases of vPSG confirmed RBD followed up prospectively for an average of 4.6 years, phenoconversion to a neurodegenerative disorder occurred at a rate of 6.3%/ year, with up to 73.5% phenoconverting when followed up to 12 years from RBD onset [23 ••]. Earlier retrospective longitudinal series of vPSG-confirmed idiopathic RBD cases have all demonstrated a high phenoconversion rate of RBD to a neurodegenerative disorder [24][25][26][27]. Schenck et al. followed up subjects to 16 years and showed the development of a neurodegenerative disorder in 80.8% of their series. A recent multimodal imaging study of RBD has proven that subjects with RBD already have established abnormalities in the peripheral autonomic nervous system and locus coeruleus, which were equivalent to that observed in PD [65].
Rather than using vPSG which is the gold standard required for RBD diagnosis, most longitudinal cohort studies in PD have instead utilized screening questionnaires such as the RBD screening questionnaire (RBDSQ) [66]. The RBDSQ has a sensitivity of 0.96 in sleep clinic cohorts, and a specificity that increases to 0.92 in unselected control groups from the general population [66]. The questionnaire has since been validated for use in PD. More recent studies using the RBDSQ to assess for the presence of RBD have shown that the prevalence of RBD in PD increases over time [31,32], but these studies differed on whether RBD was a persistent feature [31] or could fluctuate over time [32]. Earlier studies utilizing less robust methodologies such as clinical interview with the spouse [33] and an unvalidated questionnaire [34] have suggested that RBD could fluctuate over time. This is also reflected in a large meta-analysis of eight studies in newly diagnosed PD (n = 2462), which reported a wide range of reported prevalence (4.3 to 69.4%) of RBD across the different studies [67]. Only one study had utilized vPSG and reported a baseline prevalence of RBD of 25% at baseline [30••].
These earlier studies in PD which have not studied RBD with the vPSG have since been superseded by studies utilizing vPSG at multiple time points and have conclusively demonstrated that RBD, which is characterized by REM sleep without atonia (RSWA), is a persistent and progressive feature in RBD. Using vPSG, the DeNoPa study group showed that the RBD prevalence had increased from 25% at baseline to 43% at 2-year follow-up [30••] and to 52% at 6-year follow-up [28]. RSWA was independently associated with disease duration and age but not other diseaserelated factors, suggesting that RSWA could be a progression marker. The authors have also introduced the concept of REM behavioral events (RBE), where REM-associated minor motor behaviors and/or vocalizations were seen but the RSWA were below cut-off values. In their series, 31 subjects identified to have RBE at baseline developed full-blown RBD within 6 years [28]: 38% of RBE positive PD subjects converted to RBD, and 18% of PD subjects with normal REM sleep at baseline converted to RBD over 2 years [30 ••]. The prevalence of RBE had also increased over time from 50% at baseline to 63% 2 years later [30••]. The authors have proposed that RBE represents a continuum between normal REM sleep and RBD, and that RBE could be regarded as prodromal RBD.
Another group that prospectively studied 22 moderate to advanced PD subjects with RBD with vPSG performed at baseline and 3 years later also confirmed that RSWA measures increased significantly in all subjects over time [29], and was associated with worsening motor fluctuation, dyskinesias, and cognitive function over time. Findings on vPSG showing worsening of RSWA measures contrasted with self-reported frequencies of RBD, with 42% unchanged, 27% increasing, and 27% decreasing over time. Similarly, the severity of vPSG-recorded RBD behaviors showed no change in 59%, worsening in 27%, and improving in 14%. It has been suggested that vPSG-recorded RBD behaviors reflect phasic EMG activity, in contrast to RSWA measures which reflect tonic EMG activity. RSWA measures have been increasingly proposed as a reliable marker of PD progression that does not remit with the disease course [68].
In cross-sectional studies of PD, RBD is associated with a more severe disease phenotype with greater motor and non-motor burden [69]. Earlier prospective studies studied the association of RBD with hallucinations in PD, and they showed that the presence of RBD predicted the development of hallucinations [37,38] and was also associated with cognitive worsening [37] and higher mortality [37]. These preliminary findings have been subsequently confirmed in a prospective cohort of PD subjects followed up over 4 years with RBD diagnosed on vPSG [36]: half their cohort with RBD had developed dementia after 4 years, whereas PD subjects without RBD remained dementia free. The presence of RBD at baseline was also predictive of the development of hallucinations. Baseline REM sleep atonia loss was also predictive of subsequent dementia. Another large cohort study involving 923 early PD subjects followed up over an average of 4.8 years also showed that the presence of possible RBD at baseline predicted greater non-motor, motor progression over time, and was also associated with an increased risk of developing mild cognitive impairment [6]. However, another study utilizing the RBDSQ did not find an association between RBD and the subsequent development of impulse control disorders [35].
One challenge in the use of RSWA measures as a biomarker is that vPSG protocol and analysis methods used to study RBD differ across different studies. The currently accepted clinical rules of documenting RWA on vPSG using the International Classification of Sleep Disorders-3 (ICSD-3) rules do not provide cut-offs, so there is no consensus as to the EMG activity required to define RWA [70]. A standardized protocol to diagnose RBD and identify prodromal RBD has been recently proposed [71••]. Existing criteria such as the SINBAR criteria [72] to determine RWA are technically challenging, and non-informative vPSG recordings are common. This will impact upon the estimation of the true prevalence of RBD, even when performed with gold standard vPSG. The cost and night-to-night variability of RBD motor behavior severity on vPSG also places limitations on the utility of vPSG for longitudinal assessment [73,74].

Restless Legs Syndrome
RLS is characterized by unpleasant sensations producing the urge to move the legs, relief with movement, typically occurring at rest, and in late evening with a well-defined circadian rhythm [75]. Periodic limb movements of sleep (PLMS) occur in 80-89% of those with RLS [76], and can be quantified using the periodic limb movement index, which is a vPSG measure. However, the occurrence of PLMS itself is common and non-specific, and can occur with no symptoms or in association with certain medications [75]. The diagnostic criteria for RLS have undergone multiple revisions: earlier criteria were designed as a research screening tool rather than a clinical diagnostic tool and were criticized for their low positive predictive values ranging from 50 to 60%, leading to excessively high prevalence estimates of RLS in some studies [75]. There have been five studies of RLS or PLM index in PD utilizing different methods of RLS ascertainment, but only one incorporated a clinical assessment to diagnose RLS [43].
Both RLS and PD are disorders resulting from dopamine dysfunction. It has been hypothesized that RLS could be a prodromal marker for PD [77]. To date, there have been two large epidemiological studies which have reported an increased risk of PD in subjects with RLS. In the Health Professionals Follow-up Study, where subjects were studied prospectively over 8 years, male subjects experiencing RLS symptoms more than 15 times a month had a higher risk of PD development (adjusted relative risk 1.47); however, this finding was only statistically significant in PD diagnosed within 4 years of follow-up [42]. Another retrospective population study using over 8 years using the Veterans database with RLS defined according to ICD-9-CM codes showed an increased risk of PD in RLS (HR 2.57) [40].
However, RLS is an extremely common disorder, it remains unclear whether the observed association of RLS in PD is a chance phenomenon, an early symptom or late complication of PD. RLS is also linked to a myriad of other diseases common in the aging population. Trenkwalder and colleagues extensively reviewed the data associating RLS with different disorders and concluded that in the case of PD, the methodology of earlier studies were poor but that an association might be possible [78].
Two longitudinal studies, assessing previously untreated PD patients, showed that the prevalence of RLS increased over time. One study with follow-up over 26 months reported an incidence rate of RLS of 47 per 1000 case/person per year with 83.3% developing RLS within 24 months from starting dopaminergic therapy with a median latency of 7.5 months [43]. Another study followed up newly diagnosed, drugnaïve PD subjects that found an increase of RLS prevalence from 6.5% at baseline to 16.3% after 4 years [41], with the authors suggesting that RLS could occur as a complication of PD. RLS was found to be associated with higher age of PD onset, preserved dopaminergic pathways, insomnia at baseline, and EDS on follow-up. However, this study found that the baseline prevalence of RLS in PD did not differ from healthy controls and does not suggest that RLS is a prodromal feature of PD.
Only one study evaluated prognostic factors in PD progression with vPSG performed at baseline: the PLMD index did not differ between de novo PD and healthy controls, but an elevated PLMD index was shown to be a significant predictor of cognitive impairment [39].

Shift Work Disorder/Sleep Duration
The role of sleep in the clearance of amyloid from the cerebrospinal fluid via the glymphatic pathways has been recently established, which may suggest a role for sleep in the prevention of Alzheimer's disease [79••]. However, its relevance to the pathophysiology of PD remains to be confirmed. In an earlier epidemiological study using data from the US Nurses' Health Study, paradoxically, subjects who had worked night shifts for 15 years or more were shown to have a lower risk of PD compared to those who have never worked night shifts [44].

Studies Studying Multiple Sleep Disorders or Complaints
One of the difficulties in the longitudinal study of the various sleep disorders in PD is that individual sleep disorders or sleep complaints do not exist in isolation, with many sleep disturbances co-existing in the individual patient; thus, the study of individual sleep complaints in isolation may lack applicable clinical relevance. Some of the sleep complaints studied may refer to a symptom complex rather than a single sleep disorder. Furthermore, nighttime sleep disorders such as OSA and RLS that cause poor nighttime sleep quality can lead to EDS, or poor daytime sleep habits such as daytime napping that further impacts upon nighttime sleep quality.
In an epidemiological studying using data from the National Health Insurance Research Database, patients with non-apnea sleep disorders were found to have a higher risk of developing PD (crude HR 1.63) [50]. When approaching research into sleep disorders in PD, there is a dilemma whether to study individual sleep disorders, or study all causes of poor sleep in the individual. A number of longitudinal studies have studied sleep in PD using more general questionnaires designed to investigate sleep holistically in the patient; other groups have taken the quantitative approach of studying sleep using vPSG to prognosticate in PD. Sleep questionnaires and vPSG measure different aspects of sleep and both have limitations and advantages, and do not necessarily correlate with each other.
Most of the studies using general sleep scales have suggested that sleep disorders do not progress in PD or change over time [2,10,54,55], and could also improve [48]. One study found increases in the sleep/fatigue sub-score of the Non-Motor Symptoms Scale (NMSS); however, the overall effect size was very small [47]. Findings from these studies using general sleep scales are in contrast to other studies that used sleep questionnaires studying individual sleep disorders or vPSG-based studies. Although some of these scales have been recommended for PD, the use of more generic scales does not seem as helpful.
Our group took a different approach and looked at the burden of EDS, insomnia, and pRBD separately in the Parkinson's Progression Markers Initiative (PPMI) cohort to determine the burden of the respective sleep disorders at an individual patient level over 5 years [49 •]. Different sleep questionnaires were utilized to study the individual burden of the sleep disorders. Over time, sleep disturbance increased with more subjects reporting an increased number of sleep disturbances. We found that the frequency of insomnia increased the most, followed by EDS and pRBD. An earlier study using the PPMI cohort over a shorter duration of 2 years reported a significant increase in the prevalence of EDS but not sleep problems and RBD [51]. Interestingly, we also found that the number of subjects reporting multiple sleep disturbances also increased over time. After 5-year follow-up, only 30.3% of PD patients did not report any sleep disturbances, 39.0% PD reported one type of sleep disturbance, 23.4% two types of sleep disturbances, and 7.3% reported all three types of sleep disturbances. However, there was a large variability in the combination of different sleep problems in individual patients, suggesting that they might have different pathogenesis.

Predictive Values of vPSG
While vPSG is considered the gold standard required for the diagnosis of sleep disorders, including RBD, this test has also shown to have prognostic potentialities. As earlier described, vPSG features seen in RBD have prognostic value in PD: tonic REM sleep atonia loss can predict dementia [36] and this was replicated in another study [45]. However, vPSG findings have prognostic significance that is not confined to those with RBD alone. The latter study had also examined the rest of the vPSG data, but it did not find any association between other vPSG features and decline that was measured using the global deterioration scale. However, the same group studied slow wave sleep in a different cohort and found that lower N3 percentage at baseline was significantly associated with decline in MOCA scores over time [46]. Other vPSG features, including lower sleep spindle density and amplitudes at baseline, were found to be predictive of dementia in PD [53]. In particular, lower sleep spindle amplitudes in the parietal and occipital areas were correlated with poorer visuospatial abilities in demented PD subjects. The same group extended their work to show that specific vPSG features in both REM and NREM sleep and wakefulness had prognostic implications in PD [52]. Higher absolute power in the delta and theta bands in REM sleep, lower sigma power in the parietal regions in NREM sleep and lower dominant occipital frequency, and higher delta and slowing ratios during wakefulness were predictive of subsequent dementia.

Conclusion
The vast majority of prospective, longitudinal PD cohorts have instead used a variety of self-administered, subjective sleep questionnaires to study sleep disorders within a more comprehensive data collection strategy. Investigating sleep disorders is usually not the main focus of such studies, and conflicting data seen in PD studies of sleep is very likely to represent different methodologies used to determine the presence of sleep disorders. Despite these study limitations, existing longitudinal sleep disorders in PD suggest that the prevalence of sleep disorders increase over time and are associated with other disease features of PD, particularly non-motor symptoms. RBD is an established prodromal feature of PD, but it remains unclear whether other sleep disorders predict future PD. There are effective treatments for many sleep disorders, but it is unclear whether treatment of sleep disorders can alter PD symptoms and their progression or reduce the risk of future PD development. We may be underusing powerful biomarkers of brain health in neurodegeneration and failing to consider good sleep in itself as a neuroprotective therapy.
Author Contribution ZX performed the literature search and drafted the first draft. KNA drafted and critically reviewed the work. NP conceptualized the review article and critically reviewed the work.

Conflict of Interest The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.