Toward a Digital Future in Bipolar Disorder Assessment: A Systematic Review of Disruptions in the Rest-Activity Cycle as Measured by Actigraphy

Prior to the advent of actigraphy research in BD, in addition to sustained changes in emotional states, disturbed sleep and activity patterns were observed clinically to also be core features of acute episodes of BD. These patterns present differently depending on mood states and can be predictive of both onset and severity of illness (3). Daily rhythms are central to the presentation of both depression and mania, such that changes in both sleep and activity levels play a prominent role in the criteria for acute episodes of BD as outlined in the DSM-5 (2). Sleep problems present in depressive episodes in the form of insomnia or hypersomnia in the majority of cases (4), whereas hypo/manic episodes are often characterized by marked reductions in sleep accompanied by subjective reports of not feeling tired (5). Further, individuals with BD, even when euthymic, have been observed to show delayed onset of the sleep-wake phase and lower levels and later timing of melatonin secretion, an important hormone in circadian rhythm regulation (6–8). Successful chronotherapeutic interventions that modify rest-activity patterns in BD have been reported (9–11), suggesting that they constitute promising intervention targets.

Disruptions in rest-activity patterns have been previously linked to the onset of BD episodes, such that hypomania and mania has been reported following a night of sleep deprivation (12). Pinho et al. (13) found a significant relationship between disruptions in patterns of daily activity and sleep and the severity of depressive symptoms, such that the degree of disruption in patterns was associated with psychosocial dysfunction. The persistence of disruptions during euthymic periods, albeit at a lower magnitude, also suggests that this feature of BD can persist outside of syndromal mood episodes and might be a trait feature of the disorder. During euthymia, individuals with BD have shown reduced daytime activity (14, 15) and greater intra-daily variability of rest-activity patterns (15, 16) compared to healthy controls (HCs). Difficulty in falling and staying asleep (17–19) and early morning awakening persist during euthymic periods (20). Additional studies suggest that disturbed rhythms during euthymia may relate to vulnerability for future episodes and adverse outcomes as they have been associated with a greater severity of subsequent depressive and manic episodes over a 12-month period (21), a history of psychosis, and suicide attempts (22).

Biological processes that underlie circadian rhythms are implicated in the sleep-wake cycle. Quantifiable proxy markers for this cycle are rest and activity patterns, both of which, notably, can influence each other. Holistic examinations of disturbance across the 24-h cycle, not just during the rest period, (23–26), are therefore warranted. This comprehensive approach to rhythmicity is further reflected in the DSM-5 diagnostic criteria for episodes of BD which include increased energy or activity alongside decreased need for sleep in manic episodes, and psychomotor retardation and loss of energy alongside insomnia or hypersomnia in depressive episodes (2).

Sleep-specific measures and measures of the peripheral neurohormone, melatonin, have largely been employed in past studies of the sleep-wake cycle in persons with mood disorders. Polysomnography (PSG) and sleep diaries have traditionally provided objective and subjective information about sleep patterns, respectively. These methods support clinical reports of disrupted sleep patterns in BD across mood states and in comparison to both HCs and those with MDD [characterized by periods of depression in the absence of hypo/mania] (27, 28). However, these methods provide only limited data about activity during wakefulness, and while sleep diaries may also include questions about activity, they are often limited in scope and rely on subjective responses. Nevertheless, correcting melatonin abnormalities has been suggested to be the mechanism underlying the therapeutic effect of mood-stabilizing pharmacotherapies. For example, lithium and sodium valproate have been shown to reduce melatonin suppression, thus increasing overall melatonin levels (6, 29), and potentially targeting mania vulnerability (30). Outside of pharmacological interventions, chronotherapies such as bright light therapy, which is known to influence melatonin secretion, have been shown to be clinically effective in depression by advancing the phase of the sleep-wake rhythm (10, 31).

In order to better understand BD, digital technologies have been leveraged to noninvasively collect biometric parameters when individuals are in their everyday environments (32, 33). Temporal profiles of rest and activity, measured with digital actigraphy methods, hold promise for elucidating a novel phenotypic marker for BD (34, 35). Identification of a salient time-varying phenotypic marker for BD may also reveal pathophysiological mechanisms of BD, which ultimately could serve as the basis for novel detection, treatment, and prevention strategies. Prior methods used to characterize the time-varying patterns of BD, such as retrospective self-reports, are limited in their ability to provide accurate data at high levels of granularity (36). The high temporal resolution of actigraphy offers an alternative approach to capturing the time-varying characteristics of BD, and over time, may lead to advancements in early identification and treatment.

Thus, actigraphy assessment of the rest-activity cycle addresses the need for longitudinal and temporally sensitive objective assessments in individuals’ natural environment. Wristwatch-like wearable actigraphy devices assess gross motor activity using measures of acceleration speed via accelerometers, providing low cost and non-invasive rest-activity assessments (37). It is important to note that actigraphy does not measure sleep directly, but rather uses movement as a proxy for estimating sleep, although terminology used in previous publications often include the use of the word “sleep.” Parameters estimated from actigraphy data include the duration of rest or sleep periods, activity levels during rest or sleep periods, time in bed (TIB), total sleep time (TST), sleep onset latency, and daytime naps (38). Table 1 describes common actigraphy parameters in more detail. Measurements are time-stamped and made continuously, often at high measurement frequencies, for example, every 1-min. This data collection format facilitates data visualization using readily available software packages. Actigraphy has a high rate of concordance with PSG and sleep diaries in healthy populations (39–41) and in individuals with BD. For instance, similar patterns of sleep and rest quality and duration have been reported across the three measures, showing high levels of inter-correlation (42) in BD (42–45), and the ability of actigraphy to sensitively distinguish between BD depressive, manic, and euthymic episodes (46).

Notably, while actigraphy is effective in estimating rest-activity patterns, it does not provide deeper measures of sleep such as brain activity (EEG), heart rhythm (ECG), or eye movements (EOG) that are provided by PSG (47). Therefore, actigraphy cannot characterize sleep physiology and sleep architecture. Actigraphy data also requires careful interpretation as, for example, time resting in bed could be mistaken for sleep time. For that reason, referring to the period encompassing sleep as “rest,” or the “rest period,” is considered most appropriate for actigraphy measures. Furthermore, it is important to distinguish circadian rhythmicity from 24-h activity patterns. Certain actigraphic measures, such as interdaily stability (IS), relative amplitude (RA), and intradaily variability (IV), are often used erroneously to characterize endogenous circadian function (48). Similarly, L₅ onset and M₁₀ onset are also often used as markers for circadian phase timing, and changes to these variables are often erroneously concluded to be phase advances or delays. While these measures provide information about 24-h activity patterns, rhythm stability, and fragmentation (see Table 1), they must not be interpreted as markers of endogenous function. Therefore, actigraphic constructs are discussed here in the context of rest-activity patterns. But it should be noted that rest-activity patterns may serve only as a proxy measure for the sleep-wake cycle, which in turn, may serve as a proxy measure of overall circadian function. It is important to note that the sleep-wake cycle is one of a number of physiological, behavioral, and cognitive functions under partial control of the circadian system, and thus, a disruption in the rest-activity or sleep-wake cycle does not necessarily equate to underlying circadian rhythm dysfunction.

The current review aims to summarize actigraphy assessments in BD published to date, encompassing both rest and activity profiles. As a key goal is to review evidence supportive of the potential clinical significance, special focus is given to findings that help to elucidate (a) the ability of these patterns to distinguish BD from comparison groups, including HC and other clinical groups; (b) the relationship these patterns have with distinct mood states within BD and with other clinical, cognitive, and brain features of the disorder; and (c) the propensity for actigraphy to decipher the therapeutic effects of treatment interventions on mood measures in BD. We therefore address three research questions: (a) what are the rest-activity patterns associated with adult BD and how do they differ from those in HCs and other clinical comparison groups? (b) what relationship do these patterns have with other characteristics of BD, including mood, cognition, and neurobiology? (c) what effect do interventions, encompassing both pharmacological and non-pharmacological therapies, have on these patterns in BD?

Actigraphy measures have previously been employed in empirical studies of BD for at least 15 years, with a number of systematic reviews measuring their ability to capture rest-activity disturbances in those with BD [e.g., (17, 49)]. However, the clinical utility and significance of this method has not yet been assessed, and thus the translation of this method to clinical practice has not yet been widely achieved. This is increasingly important given the recent shift toward digital and longitudinal methods that are adept at assessing temporally sensitive mood disorders. Therefore, the current review aims to show how actigraphy technology can pave the way for better capturing the temporally dynamic nature of BD, and ultimately aid in the discovery of mechanisms underlying BD and more nuanced targets for detection, treatment, and prevention.

Studies were independently identified by authors PP and GdQC through manual searches of the electronic databases PubMed, MEDLINE, PsycINFO, and EMBASE. While a broad systematic method was used, following PRISMA guidelines (50), emphasis was placed on a thorough literature search of actigraphy studies in BD. To this end, the following terms were used in our primary search (further detailed in Supplementary Table 1): “actigraphy,”, “actigraph,” “actimetry,” “accelerometry,” “smart watch,” or “health watch” (separated by OR) in combination (using AND) with the terms “BD” and “bipolar” (separated by OR). Only research-grade actigraphy devices were searched for (excluding commercially available devices such as FitBit or Apple Watch) in order to ensure data reliability and validity, based on clinical specificity issues (51). Eligible papers included studies of BD samples. No lower publication date limit was applied, and the search was continued until 19th February 2021. All selected articles were published in peer-reviewed journals in English.

Information extracted from identified manuscripts were sorted into four categories: study details (author names, year of publication, and title); demographic and clinical details of subject samples [number of participants, age (mean and standard deviation), gender, diagnosis category, mood status (i.e., depressed, manic, and euthymic), and medication status]; actigraphy methodology (device name, body location, recording time or duration, and sampling/recording rate used); and main study outcomes (daily activity, rest/sleep, and 24-h patterns as outlined in Table 1). Data was independently and manually extracted by PP and GdQC by going through each article and Supplementary Material, if available. Any discrepancies were discussed, checked, and cleared by the authors.

In order to capture all relevant cross-sectional and longitudinal studies using actigraphy, BD subtypes BDI, BDII, and BD NOS or OS [(not) otherwise specified, i.e., with symptoms of BD that do not meet full diagnostic criteria] under both DSM (2) and International Classification of Diseases (52) diagnostic manuals were included. All pharmacological and non-pharmacological interventions employed in actigraphy studies of BD were included to address the third research question. Overall comparators included HCs, those with related psychiatric disorders [MDD, anxiety disorders, schizophrenia, borderline personality disorder (BPD)], insomnia, or general mental health service users. Actigraphic outcomes relating to patterns over 24-h, daytime activity, and markers of rest and nighttime period were used as outcomes.

Articles were excluded from the final list during the manual data extraction process if they aimed to validate methods (i.e., using actigraphy to validate against PSG), focused on the study of children and adolescents (defined as studies that included individuals under 18 years of age) or were solely conducted in at-risk groups, case studies, descriptions of study protocols, or genetic studies.

In order to address the avenues of investigation of the rest-activity patterns measured through actigraphy, the review was organized by three research questions. First, what are the rest-activity patterns associated with adult BD and how do they differ from those in HCs and other clinical comparison groups? Second, what relationship do these patterns have with other characteristics of BD, including mood, cognition, and neurobiology? Third, what effect do interventions, encompassing both pharmacological and non-pharmacological therapies, have on these patterns in BD?

The database search identified 114 unique articles, of which 70 met systematic review criteria (Figure 1). Articles that were excluded from the final list fell into the following categories: methods validation (n = 10), the study of children and adolescents (n = 12) or at-risk groups (n = 12), case studies (n = 7), study protocols (n = 1), or genetic studies (n = 2). To address the three research questions: 41 of the 70 included articles compared actigraphy in BD with that of HCs or other clinical groups; 19 investigated the relationship between rest-activity patterns and different features of BD; and 10 articles assessed the effect of interventions on actigraphy measures in BD. Study characteristics are presented in Supplementary Table 2. All rest and activity pattern findings reported in section “Results” are derived from the 70 actigraphy studies.

As demonstrated in Supplementary Table 2, actigraphic recording periods varied greatly such that studies collected data from anywhere between 24 h [e.g., (53–55)] and 90 days (56). While 44% (n = 32) of studies accounted for the 24-h nature of rest-activity patterns, a further 39% (n = 28) focused on recordings during the rest or nighttime period only, and 17% (n = 12) focused on those of daytime activity only.

Given the small number of studies and methodological heterogeneity in the studies identified (as evidenced by varying sample sizes, length of actigraphic recording, devices used, and variables extracted), we did not calculate effect measures for each outcome. The studies were not directly comparable enough to mathematically assess meaningful differences between them. Our results below are based on the findings of each of the studies included in this systematic review.

Of 41 articles addressing this question, 49% (n = 20) directly compared rest-activity patterns in persons with BD with those of HCs and 37% (n = 15) included other psychiatric clinical comparison groups, 5% (n = 2) included persons with a diagnosis of insomnia, 2% (n = 1) included persons with childhood attention deficit hyperactivity disorder (ADHD), and 7% (n = 3) included at-risk groups [i.e., non-affected siblings or first or second degree relatives, and those with a previous episode of major depression with sub-threshold mania symptoms (57)] in addition to BD and HC groups.

Nineteen of the 70 included articles addressed this research question. Of these articles, 43% (n = 8) addressed the relationships between rest-activity patterns and specific mood states, 21% (n = 4) investigated the neuroimaging correlates of these patterns, as measured by magnetic resonance imaging, 10% (n = 2) studied the relationship with suicidal thoughts and behaviors, 10% (n = 2) assessed the association with cognition, 6% (n = 1) investigated the relationship with metabolic symptoms, and an additional 10% (n = 2) assessed the relationship between these patterns and coffee, tobacco, and alcohol consumption (85), and characteristics of social support and strain (86).

Ten of the 70 included articles addressed this research question. Of these articles, 60% (n = 6) investigated differences in rest-activity patterns after pharmacological treatment including mood-stabilizing medications (lithium and quetiapine), and 40% (n = 4) investigated differences in these patterns after non-pharmacological treatments including light therapy, cognitive behavioral therapy (CBT), and blue-blocking (BB) glasses. Risk of bias assessment (105, 106) was completed on these interventional studies and can be found in Supplementary Figure 1.

Results from this systematic review provide evidence that BD is associated with disruptions in rest-activity patterns including longer sleep duration, longer sleep onset latency (58–63), and lower average daily activity (16, 75) that may be especially associated with depression. BD mania is instead linked to more complex and variable patterns over a shorter temporal scale that can also be predictive of future relapse (54). Different mood episode types are also associated with different profiles of phase-shifting, such that a phase delay is more commonly associated in BD-depression and a phase advance in BD-mania (79). Disruptions in rest-activity patterns are shown to persist during euthymia [e.g., (14, 16, 89)] and there is preliminary evidence that distinct patterns may differentiate BD from other major mood and psychotic disorders, and within subtypes of BD (7, 69, 81). Both pharmacological and non-pharmacological therapies have demonstrated rest-activity pattern stabilization associated with improvements in mood symptoms [e.g., (107, 113)]. However, the mechanisms by which these occur are yet to be fully uncovered. This may be further facilitated by technological and digital advancements provided by actigraphy, allowing the collection of fine-grained, longitudinal, and temporally sensitive parameters while individuals are in their typical environments, and emerging machine learning and other computational strategies to extract maximal information from time series data. The extant literature supports the promise of actigraphy to pave the way for the development of improved early detection, treatment, and prevention methods.

It is important, however, for current limitations to be addressed, including variability in methodology (e.g., the actigraphic variables chosen for analyses, actigraphic recording periods, and small sample sizes) and study sample characteristics (e.g., medication use and age composition) which can impede further meta-analytic work (122). Additionally, future studies should account for variability in daily patterns due to societal demands (e.g., shift work), in order to consider how such circadian misalignment may lead to artificially fluctuating levels of actigraphic fragmentation (123) that are not representative of rest-activity patterns or endogenous circadian rhythms (124). While meta-analytic work could provide a statistical summary of results, the disparities in study characteristics did not allow for this type of analysis. It is also important to note the inherent limitations in systematic review methodology. This systematic review might exclude relevant literature as, inherent in a systematic review approach, it is bounded by search terms. Of note is the recent expansion in health-related digital technologies available for use on both the research and personal-use market. While we chose not to include commercially available digital technologies in our current review due to their limited specificity and accuracy in reporting rest-activity patterns compared to research-grade devices (51), future research could expand on this as the devices become more ubiquitous in the general population and advancements are made in optimizing their research and clinically oriented utility.

Future actigraphy research in BD could not only benefit the work of researchers aiming to further elucidate the neurobiological basis of BD, but also clinicians and patients, who may in the future be able to use real-time and longitudinal monitoring to identify personal rest-activity patterns and warning signs (35, 125). Traditional modes of in-person clinical appointments at the temporal scale of weeks or months can benefit from the use of daily and on-going monitoring through digital technologies and passive activity sensors for their ability to provide longitudinal and prospective data, including interim data that may indicate need for more acute care (34). The use of digital technologies, such as smartphone-based ecological momentary assessment in combination with actigraphy, is gaining traction for its ability to provide complementary information on day-to-day fluctuations in mood and other symptoms, social and other daily functioning, as well as changes in relation to daily stressors and life events (122, 126–128). Further, given the expedited need for remote and digital assessment tools and telehealth provisions that have intensified due to the ongoing COVID-19 pandemic and its associated increase in mood symptoms and disorders (129–131), particularly BD, the additional value of longitudinal monitoring of rest-activity rhythms and its potential to flag possible preceding events for relapse and worsening of symptoms is immense (132–134).

As rest-activity changes can be robust in precipitating mood episodes, and are possible early indicators of impending episodes and of the beneficial effects of treatments, interventions targeted at regulating rest-activity patterns and those that integrate actigraphy over time may be especially effective in treatment and prevention strategies. Interpersonal and Social Rhythm Therapy (IPSRT), which has an SRT component designed to help individuals regularize their rest-activity pattern, is one of the few psychotherapies shown in clinical trials to be effective in treating BD (135–142). The rhythm regularization of IPSRT has been shown to reduce risk of recurrence over 2 years (143) with changes in regularity of daily routines mediating symptomatic outcomes, supporting rhythm regulation as a treatment target with potential for sustained benefits and prognosis improvements. Studies are underway with modified versions of SRT (144), for example a version delivered largely via telehealth that has shown promise in reducing mood symptoms and suicide propensity in BD (134).

Combining digital monitoring, and holistic multi-modal assessment of dynamic symptomatology, with neuroimaging and cognitive assessments can help enable the field to uncover the neurobiological mechanisms responsible for the phenotypic presentations of BD. Larger scale studies will be especially helpful in elucidating heterogeneous aspects. As can be seen in Supplementary Table 2, several actigraphy devices are available and used for research purposes in clinical populations. Although these devices all estimate body movement through accelerometers on the x, y, and z planes, there can be heterogeneity in the types of actigraphy parameters extracted from the raw data. Table 1 outlines several different variables that can be referred to, e.g., when reporting disruptions in rest (e.g., SE, WASO, L₅ activity). The field has yet to reach consensus on which parameters are key for our understanding of rest-activity patterns in BD. Going forward, optimal parameters and modalities should be further operationalized to ensure consistency in reporting across studies and to allow for the combined analysis of datasets. It should also be noted that estimating sleep via proxy markers, such as L₅, can be problematic because of its derivative nature (73). Collaborative efforts such as those of the Motor Activity Research Network for Health (mMARCH) will also facilitate efforts to use common procedures and methods for data extraction and analysis. For example, a processing pipeline for actigraphy based on the GGIR package that was developed for mMARCH sites (145) extracts features of sleep, physical activity, and circadian rhythmicity, and applies JIVE to capture the joint variation across three domains. Consideration of the joint variation will lead to a better understanding of the interrelationships of rest, activity, and their rhythms within and between days. This package also includes standard methods for handling missing data and non-wear time that has not been systematically presented in most studies of actigraphy in BD. Variability of the features such as IV and IS, which capture regularity in activity patterns within and across days, respectively, as well as circadian features, will be particularly relevant to future studies of BD. These markers are key for their ability to address both rest and activity, as well as highlight the importance of variability across time – two aspects proposed to be fundamental to BD. Further, novel methods to assess this variability are still being developed and could be an important future methodological direction for study in this field (123). For full characterization of sleep, future studies should also use PSG and/or other well-established and more comprehensive sleep measures, as well as capture longer study periods and greater sample sizes.

The original contributions presented in the study are included in the article/, further inquiries can be directed to the corresponding author/s. Supplementary Material