(C) PLOS One [1]. This unaltered content originally appeared in journals.plosone.org.
Licensed under Creative Commons Attribution (CC BY) license.
url:https://journals.plos.org/plosone/s/licenses-and-copyright

------------



Association of accelerometer-derived sleep measures with lifetime psychiatric diagnoses: A cross-sectional study of 89,205 participants from the UK Biobank

['Michael Wainberg', 'Krembil Centre For Neuroinformatics', 'Centre For Addiction', 'Mental Health', 'Toronto', 'Samuel E. Jones', 'Institute For Molecular Medicine Finland', 'Fimm', 'University Of Helsinki', 'Helsinki']

Date: 2021-10

Abstract Background Sleep problems are both symptoms of and modifiable risk factors for many psychiatric disorders. Wrist-worn accelerometers enable objective measurement of sleep at scale. Here, we aimed to examine the association of accelerometer-derived sleep measures with psychiatric diagnoses and polygenic risk scores in a large community-based cohort. Methods and findings In this post hoc cross-sectional analysis of the UK Biobank cohort, 10 interpretable sleep measures—bedtime, wake-up time, sleep duration, wake after sleep onset, sleep efficiency, number of awakenings, duration of longest sleep bout, number of naps, and variability in bedtime and sleep duration—were derived from 7-day accelerometry recordings across 89,205 participants (aged 43 to 79, 56% female, 97% self-reported white) taken between 2013 and 2015. These measures were examined for association with lifetime inpatient diagnoses of major depressive disorder, anxiety disorders, bipolar disorder/mania, and schizophrenia spectrum disorders from any time before the date of accelerometry, as well as polygenic risk scores for major depression, bipolar disorder, and schizophrenia. Covariates consisted of age and season at the time of the accelerometry recording, sex, Townsend deprivation index (an indicator of socioeconomic status), and the top 10 genotype principal components. We found that sleep pattern differences were ubiquitous across diagnoses: each diagnosis was associated with a median of 8.5 of the 10 accelerometer-derived sleep measures, with measures of sleep quality (for instance, sleep efficiency) generally more affected than mere sleep duration. Effect sizes were generally small: for instance, the largest magnitude effect size across the 4 diagnoses was β = −0.11 (95% confidence interval −0.13 to −0.10, p = 3 × 10−56, FDR = 6 × 10−55) for the association between lifetime inpatient major depressive disorder diagnosis and sleep efficiency. Associations largely replicated across ancestries and sexes, and accelerometry-derived measures were concordant with self-reported sleep properties. Limitations include the use of accelerometer-based sleep measurement and the time lag between psychiatric diagnoses and accelerometry. Conclusions In this study, we observed that sleep pattern differences are a transdiagnostic feature of individuals with lifetime mental illness, suggesting that they should be considered regardless of diagnosis. Accelerometry provides a scalable way to objectively measure sleep properties in psychiatric clinical research and practice, even across tens of thousands of individuals.

Author summary Why was this study done? Sleep problems are both symptoms of and risk factors for many mental health conditions.

This study aimed to determine how objectively measured sleep differs among individuals with lifetime psychiatric diagnoses. What did the researchers do and find? This cohort study of 89,205 individuals from the UK Biobank analyzed 10 accelerometer-derived sleep measures.

The study found a rich suite of associations with lifetime diagnoses of psychopathology and psychiatric polygenic risk scores, though effect sizes were generally small. What do these findings mean? Sleep pattern differences are the norm among patients with lifetime psychiatric illness.

Accelerometry provides a scalable way to objectively measure such differences in psychiatric research and practice.

Limitations include the use of accelerometer-based sleep measurement and the time lag between psychiatric diagnoses and accelerometry.

Citation: Wainberg M, Jones SE, Beaupre LM, Hill SL, Felsky D, Rivas MA, et al. (2021) Association of accelerometer-derived sleep measures with lifetime psychiatric diagnoses: A cross-sectional study of 89,205 participants from the UK Biobank. PLoS Med 18(10): e1003782. https://doi.org/10.1371/journal.pmed.1003782 Academic Editor: Vikram Patel, Harvard Medical School, UNITED STATES Received: May 4, 2021; Accepted: August 25, 2021; Published: October 12, 2021 Copyright: © 2021 Wainberg et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: De-identified data for the 10 accelerometer-derived sleep measures used in the study are available through the UK Biobank. The data are available to researchers through a procedure described at http://www.ukbiobank.ac.uk/using-the-resource/. Funding: The authors acknowledge Milos Milic for data curation assistance. MW and SJT acknowledge support from the Kavli Foundation, Krembil Foundation, CAMH Discovery Fund, the McLaughlin Foundation, NSERC (RGPIN-2020-05834 and DGECR-2020-00048) and CIHR (NGN-171423). DF is supported by the Michael and Sonja Koerner Foundation New Scientist Program, Krembil Foundation, CAMH Discovery Fund, and the McLaughlin Foundation. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. This research was conducted under the auspices of UK Biobank application 61530, “Multimodal subtyping of mental illness across the adult lifespan through integration of multi-scale whole-person phenotypes”. Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: M.A.R. is on the SAB of 54Gene, Related Sciences and scientific founder of Broadwing Bio and has advised BioMarin, Third Rock Ventures and MazeTx; the remaining authors declare no competing interests. Abbreviations: FDR, false discovery rate; HDCZA, Heuristic algorithm looking at Distribution of Change in Z-Angle; ICD, International Classification of Diseases; LD, linkage disequilibrium; REM, rapid eye movement; WASO, wake after sleep onset

Introduction Sleep is fundamental to mental health. Poor sleep is not just a hallmark of psychiatric disorders, but can be a causal risk factor as well [1]. Sleep interventions can lessen depression [2] and posttraumatic stress disorder [3] symptoms, prevent psychotic experiences [4,5], and improve psychological well-being and quality of life [6]. In psychiatry, sleep properties are often ascertained via self-report: for instance, self-reported sleep quality is a component of nearly every depression rating scale, including the HAM-D [7] and Montgomery–Asberg [8]. However, self-reported measures of sleep do not always correlate well with direct physiological measurements: prior work has found that a typical person may overestimate [9,10] or underestimate [11,12] their sleep duration by up to 75 minutes, relative to direct measurement. This divergence may be especially large among psychiatric patients: individuals with depression are less accurate at reporting sleep quality and duration than healthy controls [13]. Thus, when studying sleep in a psychiatric context, objective measurement may be a useful complement to self-report. While lab-based polysomnography remains the gold standard for sleep measurement, it is ill-suited to long-term or home use, and spending a night in a sleep clinic with multiple electrodes attached to one’s body may not be conducive to a good night’s sleep. Wrist-based accelerometry (also called actigraphy) is a reasonably accurate and much more versatile and scalable alternative [14–19]. Historically, most accelerometry studies of sleep and mental illness have relied on highly selected samples of tens to hundreds of individuals [20]. Recently, the UK Biobank collected 7-day accelerometry recordings from over 100,000 participants [21], providing an unprecedented opportunity to study the interplay between sleep and mental health across a broad cross-section of the community. Researchers have used this dataset to determine that circadian dysrhythmia is correlated with mood disorders and subjective well-being [20] and genetically correlated with mood instability [22] and that insomnia, chronotype [23], sleep duration [24], and daytime sleepiness [25] are genetically correlated with lifetime prevalence of several psychiatric disorders. Yet despite recognition that insomnia and disturbed sleep are transdiagnostic processes [26,27] that cut across conventional diagnostic boundaries, the relationship between objectively measured sleep and mental health has rarely been studied from a transdiagnostic perspective—and even then, often only for a single sleep property at a time and in a small sample. To illustrate this research gap, we searched PubMed for studies of objectively measured sleep in a psychiatric context, using the search terms “sleep AND (polysomnography OR accelerometry OR actigraphy) AND (depression OR anxiety OR bipolar OR schizophrenia),” and identified 2,923 articles meeting these criteria. However, after narrowing our search criteria to studies considering all 4 disorders—“sleep AND (polysomnography OR accelerometry OR actigraphy) AND (depression AND anxiety AND bipolar AND schizophrenia)”—we identified only 4 articles: 2 reviews [28,29], a case series of 58 patients [30], and a cohort study of 110 patients also focused on sleep apnea [31]. Here, we address this research gap by performing an “all-by-all” analysis of sleep and mental health across 89,205 UK Biobank participants. Specifically, we investigate the associations of 10 sleep measures—including bedtime and wake-up time, sleep duration, number of awakenings, and variability in bedtime and sleep duration—with 4 lifetime psychiatric diagnoses—major depressive disorder, anxiety disorders, bipolar disorder/mania, and schizophrenia spectrum disorders—as well as polygenic risk scores for major depression, bipolar disorder, and schizophrenia.

Methods This study is reported as per the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (S1 Checklist). The study did not have a prospective protocol or analysis plan. Cohort Accelerometry recordings were gathered from 103,688 participants in the UK Biobank, a community-based prospective cohort study, between 2013 and 2015 [21]. Briefly, participants were provided with an Axivity AX3 triaxial accelerometer by mail and asked to wear it on their dominant wrist for 7 days, starting immediately after receiving it in the mail. These data have been made available as Data-Field 90001 of the UK Biobank (“Acceleration data—cwa format”). Of these 103,688, participants were excluded if they did not wear the accelerometer for every one of the 24 hours in a day on at least one of the days (Data-Field 90084, “Unique hours of wear in a 24 hour cycle (scattered over multiple days)”; N = 4,345); if their accelerometer was not well calibrated (Data-Field 90016, “Data quality, good calibration”; N = 11); if their wear period included a DST change (Data-Field 90018, “Daylight savings crossover”; N = 4,543); if they woke up in the afternoon on an average day (for instance, shift workers; N = 137); or if fewer than 2 days during the 7-day wear period were valid (see below; N = 6,020). Due to the inclusion of analyses involving polygenic risk scores, participants were also excluded if they had greater than 2% genotype missingness (Data-Field 22005, “Missingness”), a mismatch between genetic sex and self-reported sex, sex chromosome aneuploidy, or were flagged as “Outliers for heterozygosity or missing rate” (Data-Field 22027). Self-reported white participants (according to Data-Field 21000, “Ethnic background”; N = 86,513) were used for the primary analysis, with replication in a much smaller number of self-reported non-white participants (N = 2,692), for a total of 89,205 participants. Replication was also performed stratified by sex, among self-reported white females (N = 48,562) and males (N = 37,951). Accelerometry data processing Accelerometry recordings were temporally segmented into sleep and activity bouts using an accelerometry software toolkit (github.com/activityMonitoring/biobankAccelerometerAnalysis) specifically designed for the UK Biobank [32,33]. As described previously, this segmentation was performed by a machine learning classifier consisting of a random forest, the predictions of which are temporally smoothed by a hidden Markov model. This classifier was trained on an external, labeled dataset of accelerometer recordings. For our analyses, we ignored distinctions between activities and classified each bout as either “sleep” or “wake.” Bouts for times when the accelerometer was not worn were probabilistically imputed; we labeled these bouts as “sleep” if the imputed probability of sleep was greater than 0.5, and “wake” otherwise. While this segmentation is sufficient to determine the start and end time of each sleep and wake bout, it does not annotate which bouts make up the primary sleep period (usually at night) and which are just naps. To do this, we used steps 7 to 10 of the Heuristic algorithm looking at Distribution of Change in Z-Angle (HDCZA) algorithm implemented in the widely used GGIR accelerometry toolkit [34]: following GGIR, we defined each day’s primary sleep period as the longest time period containing sleep bouts of at least 30 minutes separated by gaps of no more than 60 minutes. (While this definition is commonly used in the field, there is no single correct definition of what should constitute sleep inside versus outside the primary sleep period, particularly for individuals with highly fragmented sleep.) A “day” was defined as the period from 3 PM to the following 3 PM. Days were deemed invalid and discarded if their primary sleep period crossed one of the 3 PM day boundaries, if all the day’s sleep periods were less than 30 minutes, or if more than 10% of the day’s data was imputed. Having defined each day’s primary sleep period, we defined 10 sleep measures based on the timings and lengths of the sleep and wake bouts inside and outside of this period (Table 1). These measures are similar to those used in previous accelerometry and polysomnography studies [35,36]. All measures were quantified as medians (or median absolute deviations, for the variability measures) across days, to be robust to outliers. To keep the focus on sleep, we do not include activity features, nor the L5 and M10 measures of circadian rhythmicity used in a previous study of the UK Biobank [20], which are based on both sleep and activity. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Table 1. The 10 sleep features and their definitions. Medians and mean absolute deviations are taken across all valid days. https://doi.org/10.1371/journal.pmed.1003782.t001 Inpatient psychiatric diagnoses These 10 sleep measures were tested for association with 4 lifetime inpatient psychiatric diagnoses from any time before the date of accelerometry: schizophrenia spectrum disorders (International Classification of Diseases [ICD] codes F20-F29), bipolar disorder/mania (F30, F31), major depressive disorder (F32, F33), and anxiety disorders (F40, F41). Inpatient diagnoses and their dates were derived from the “hesin_diag” table of the inpatient records provided by the UK Biobank (Data-Field #41234, “Records in HES inpatient diagnoses dataset”). To mitigate contamination of the control group, we excluded participants with preexisting primary care diagnoses (available for approximately 45% of the cohort), death record-based diagnoses, and/or self-reported clinician diagnoses of the same disorder, according to the “Source of report of [ICD code]” fields provided with the UK Biobank, for instance, Data-Field 130895, “Source of report of F32 (depressive episode).” We also excluded participants whose first inpatient diagnosis of the disorder was after the date of accelerometry. For instance, when computing associations with inpatient major depressive disorder, we excluded participants with primary care, death record-based, or self-reported major depressive disorder diagnoses, or whose first inpatient diagnosis of major depressive disorder was after the date of accelerometry. Polygenic risk scores The 10 sleep measures were also associated with polygenic risk scores derived from public genome-wide association study results for major depression [37], bipolar disorder [38,39], and schizophrenia [40] across self-reported white participants. The UK Biobank’s imputed genotypes were filtered using version 2.0 of the plink software [41]. Nonautosomal variants, duplicates, indels, and variants with imputation INFO score less than 0.8 were removed, as were variants with Hardy–Weinberg equilibrium p-value less than 10−10, over 5% missingness, minor allele frequency below 0.1% across self-reported white participants. The polygenic risk scores were then calculated. Summary statistics were first harmonized with the UK Biobank imputed genotypes with respect to reference/alternate allele and strand, using the allele harmonization framework from munge_sumstats.py in the ldsc software package [42]. Ambiguous variants (A/T, C/G, G/C, T/A) and variants missing from UK Biobank were excluded. Summary statistics were then subset to p < 0.05, a threshold found to be most predictive across self-reported white participants in the UK Biobank (S1 Fig). Frequency-informed linkage disequilibrium (LD) pruning to r2 > 0.2 across the self-reported white participants was then performed using a 500-kb sliding window. The remaining variants constituted the trait’s polygenic risk score, with the variants’ effect sizes (β coefficients for educational attainment, log odds ratios for the other 3 case–control studies) constituting the weights of the score. Finally, polygenic risk scores were scored on each individual in the study cohort by summing, across the variants in the polygenic risk score, the variant’s weight times the individual’s number of effect alleles of that variant; missing genotypes were mean imputed. Association analyses Association tests were performed by linearly regressing the outcome variable (sleep measures) on the exposure variable (psychiatric diagnoses or polygenic risk scores). Covariates consisted of age and season at the time of the accelerometry recording, sex, Townsend deprivation index (an indicator of socioeconomic status), and the top 10 genotype principal components. Benjamini–Hochberg correction [43] was performed at a false discovery rate (FDR) threshold of 5%. Analyses of self-reported sleep properties As a secondary analysis, we considered 6 self-reported sleep properties (S1 Table) ascertained at baseline assessment between 2006 and 2010, approximately a half decade earlier than the accelerometry. We first assessed the concordance between self-reported sleep properties and accelerometry-derived sleep measures, by linearly regressing each accelerometry-derived measure (as the dependent variable) on each self-reported sleep property (as the independent variable) across all 77,232 self-reported white participants with both types of sleep properties, using the same covariates as above. Next, we performed the same battery of associations with psychiatric diagnoses and polygenic risk scores, with the following differences from the primary analysis. First, we analyzed all 400,771 self-reported white participants with self-reported sleep properties and genotype data, not just the 89,205 with accelerometry. Second, we excluded participants with inpatient diagnoses after the baseline assessment, rather than after the date of accelerometry. Third, instead of including the age and season of accelerometry as covariates, we include the age at baseline assessment. Aside from these changes, this secondary analysis was conducted identically to the primary analysis. Ethics statement This study is a reanalysis of the UK Biobank cohort, which obtained ethical approval and informed consent from study participants as described in the flagship UK Biobank publication [44]. This study was conducted under the auspices of UK Biobank application 61530, “Multimodal subtyping of mental illness across the adult lifespan through integration of multiscale whole-person phenotypes.”

Discussion In this work, we analyzed the structure of sleep and its association with lifetime psychopathology across nearly 90,000 individuals. In a departure from previous studies analyzing only a single sleep property or a single disorder, we take an “all-by-all” approach, associating 10 accelerometer-derived sleep measures with 4 inpatient psychiatric diagnoses and 3 psychiatric polygenic risk scores. On the whole, accelerometer-derived sleep measures were concordant with self-reported sleep properties, and both were richly associated with psychiatric diagnoses and polygenic risk scores, and these associations replicated across ancestries and sexes. To our knowledge, this is the first large-scale transdiagnostic study of objectively measured sleep and mental health. The same sleep pattern differences tended to recur across disorders: each diagnosis was associated with a median of 8.5 of the 10 sleep measures, almost always in the direction of worse sleep quality. However, effect sizes were generally quite small. Note that these numbers are with respect to lifetime diagnoses; the extent of sleep disruption would presumably be greater during an active episode of depression, mania, or psychosis [19]. Across diagnoses, metrics pertaining to sleep quality were more strongly associated than mere sleep duration. Strikingly, the accelerometry-defined duration of an individual’s longest sleep bout was much more strongly associated with most psychiatric diagnoses and polygenic risk scores than total sleep duration. Given the intimate relationship between sleep bout length and sleep quality [46,47], this suggests that sleep quality may be more disturbed than sleep length across psychopathologies. These findings undergird the importance of assessment of sleep quality in addition to sleep duration. However, we note that effects on sleep may vary greatly across disease subtypes (for instance, atypical versus nonatypical depression) or states (for instance, manic episode versus depressive episode versus euthymia), and these effects may be obscured when lumping together subtypes and states, as we do here. Most prior studies of sleep and mental illness have focused on white individuals, and a key differentiating factor of our work is its replication across diverse ancestries, including those historically underrepresented in medical research [48]. In addition to this trans-ethnic replication, we also confirm that males and females display similar sleep alterations across lifetime psychopathologies. Even so, our results should be interpreted in the context of the UK Biobank’s well-characterized “healthy volunteer” selection bias [49] and its consequent underascertainment of individuals with psychiatric diagnoses [50]. This study has several key limitations. First, it relies on linked inpatient medical records, which may not capture all participants with clinically significant psychopathology, thus compounding the “healthy volunteer” bias mentioned in the previous paragraph. Second, the (often years-long) time lag between psychiatric diagnoses and accelerometry (Table 2) obscures whether participants were in an active manic, depressive, or psychotic episode at the time of their accelerometry. Third, the study’s cross-sectional design limits the ability to make inferences about causality. Fourth, accelerometer-based sleep measurement is not as precise as polysomnography, the gold standard in sleep research. The algorithm used for sleep/wake segmentation [32,33] was trained on accelerometry data annotated from head-mounted video and sleep diaries, rather than direct measures of sleep/wake, which could result in the misclassification of certain awake-in-bed periods (for instance, short awakenings or periods prior to sleep onset where the individual is motionless) as sleep. This may also account for the relatively high median sleep efficiency, low wake time after sleep onset, and long sleep bout durations seen in this study relative to polysomnography-based studies [51]. Also, accelerometry alone cannot accurately distinguish between rapid eye movement (REM) sleep and the various stages of non-REM sleep [52,53]. However, these limitations should be weighted against the population-scale, pan-diagnostic scope that accelerometry-based sleep measurement enables. Moreover, certain of our sleep metrics may indirectly capture aspects of sleep stage: for instance, high numbers of awakenings or low duration of longest sleep bout may indicate insufficient REM sleep [46,47,54]. A key clinical implication of this work is that sleep pattern differences are a transdiagnostic feature of psychopathology. Alterations in sleep parameters—particularly those impacting sleep quality and not merely duration—should be considered regardless of which psychiatric conditions a patient presents with. Future transdiagnostic studies of sleep and psychopathology should employ a longitudinal design to more precisely examine how sleep parameters vary across phases of mental illness. In sum, we find that alterations in objectively measured sleep parameters are the norm among patients with lifetime psychiatric illness. Our findings provide a rich clinical portrait of the ways in which sleep can be disrupted across individuals with lifetime mental illness. This work showcases the capacity of accelerometry to provide detailed, objective sleep measurements at scale, even across cohorts of tens of thousands of individuals.

Acknowledgments The authors acknowledge Milos Milic for data curation assistance. This research was conducted under the auspices of UK Biobank application 61530, “Multimodal subtyping of mental illness across the adult lifespan through integration of multi-scale whole-person phenotypes.”

[END]

[1] Url: https://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1003782

(C) Plos One. "Accelerating the publication of peer-reviewed science."
Licensed under Creative Commons Attribution (CC BY 4.0)
URL: https://creativecommons.org/licenses/by/4.0/


via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.