LASSO identified the sleep metrics or combination that best predicted demographic and clinical variables

To date these traits have only been studied in smaller samples but this approach will be invaluable as sample sizes increase. Another challenge for AUD genetics is that AUD is a dynamic phenotype, even more so than other psychiatric conditions, and therefore may necessitate yet larger sample sizes. Ever-larger studies, particularly those extending mere alcohol consumption phenotypes, are required to find the genetic variants that contribute towards the transition from normative alcohol use to misuse, and development of AUD. Furthermore, genetic risk unfolds across development, particularly during adolescence, when drug experimentation is more prominent and when the brain is most vulnerable to the deleterious effects of alcohol . The Adolescent Brain Cognitive Development , with neuroimaging, genotyping and extensive longitudinal phenotypic information including alcohol use behaviors , offers new avenues for research, namely to understand how genetic risk interacts with the environment across critical developmental windows. Population biobanks aligning genotype data from thousands of individuals to electronic health records are also promising emerging platforms to accelerate AUD genetic research . Despite these caveats, the GWAS described in Table 1 have already vastly expanded our understanding of the genetic architecture of alcohol use behaviors. It is evident that alcohol use behaviors, like all complex traits, are highly polygenic . The proportion of variance explained by genetic variants on GWAS chips ranges from 4 to 13% . It is possible that a significant portion of the heritability can be explained by SNPs not tagged by GWAS chips, indoor grow trays including rare variants . For instance, a recent study showed that rare variants explained 1-2% of phenotypic variance and 11-18% of total SNP heritability of substance use phenotypes .

Nonetheless, rare variants are often not analyzed when calculating SNP heritability, which can lead to an underestimate of polygenic effects, as well as missing biologically relevant contributions for post-GWAS analyses . Equally important is the need to include other sources of -omics data when interpreting genetic findings, and the need to increase population diversity . Therefore, a multifaceted approach targeting both rare and common variation, including functional data, and assembling much larger datasets for meta-analyses in ethnically diverse populations, is critical for identifying the key genes and pathways important in AUD. In recent years, there has been an explosion in the availability of consumer wearables which are able, through accelerometry, heart rate, and “real time” physiologic signal, to provide insights on lifestyle, including sleep. The ubiquity of these devices means that a wealth of data is now available for understanding sleep in diverse populations, including those with serious mental illness such as bipolar disorder . Sleep disturbance is highly prevalent in BD and associated with psychopathology . Using objective forms of sleep measurement in BD may be more accurate than clinical interviews, as self-report of sleep may be influenced not only by recall, but by current affective states inherent in BD . While polysomnography is the most validated means of measuring sleep, wearables can generate important data about sleep quality that has been validated against polysomnography . Indeed, past studies show altered sleep patterns as measured by actigraphy in BD vs. healthy subjects as example. Accelerometry can estimate a number of sleep parameters, including total sleep time , wake after sleep onset , percent sleep , number of awakenings amongst others . While these individual parameters, by themselves, are clinically relevant and provide important data to identify mechanisms for outcomes, there are some statistical challenges, including Type 1 error which can be problematic when looking at multiple outcomes as are commonly measured in studies of BD . Distilling sleep parameters into one joint construct may help identify those with poor sleep and in need of sleep intervention. Approaches such as machine learning can cluster people based upon patterns of sleep, and past studies in sleep using these methods have shown promising results .

However, these approaches have limitations in their applicability and generalizability to individual patients , and are often hard to communicate to those unfamiliar with these methods, limiting their clinical utility. Other investigators have sought to identify actigraphic measures most important in distinguishing sleep quality across subjects. For example, Natale et al. used linear discriminant function analysis to find that TST, sleep onset latency, and NA were the best combination of actigraphy statistics differentiating those with insomnia from those with healthy sleep patterns. In neuropsychological assessment, cognitive test scores are commonly standardized as compared to a normative population, and can then be merged into a composite. Using a similar approach, but with actigraphic sleep measurements, may yield a clinically meaningful way to summarize sleep quality across measures among those with BD. Sleep quality is associated with a number of demographic and clinical characteristics. For example, in community-based samples, poor sleep is associated with lower overall health quality , increased risk for depression , higher body mass index , lower cognition , higher levels of inflammation , among other negative health outcomes. In BD, poor sleep is associated with these same characteristics but at a more pronounced level; and poor sleep is also associated with worsening BD symptoms , poorer overall mood , impaired cognition , and greater inflammation , among others. To have construct validity and be clinically meaningful, a composite sleep measure in BD should be correlated with demographic and clinical characteristics consistent with past studies of sleep quality. In this study we created a composite score for sleep across accelerometer-derived sleep variables in a sample of those with BD. The main aims of this study were to identify the clinical utility of this composite measure by examining demographic, clinical, and biological correlates within subjects, and to identify the sleep variables most contributing to associations between the composite measure and these associated variables. We hypothesized that better sleep quality as seen in our composite measure would be associated with fewer depressive and mania symptoms,vertical grow racks for sale greater medication load, and lower cognition and inflammation.

We hypothesized that any observed correlations would be driven by multiple sleep indices included in the overall composite score.Data came from a longitudinal study of cognition and inflammation in BD. We recruited those with BD and HCs from outpatient clinics, community settings, and other research studies at UC San Diego. BD was defined as a diagnosis of Bipolar I or II DSM-IV Disorder receiving outpatient care. Exclusion criteria included: acute illness or pregnancy, a recent vaccination, history of various health conditions , cancer treatment in past, diabetes/hypertension that is not controlled, among others. Among HCs, we also excluded individuals who had a history of DSM-IV Axis I disorders, previous use of psychotropic medications, as well as having a first degree relative with history of depression, BD, or schizophrenia. Each year of the study, participants were asked to complete a 2-week burst of assessments consisting of three in-person visits and completion of up to 14 24-hour periods of wrist-worn actigraphy. We focused on a subset of 51 persons with BD and a comparable group of healthy controls [HCs] who had valid wrist actigraphy data which we defined as having actigraphy data recorded for at least five nights. Of note, some individuals had more than one period of actigraphy assessment or did not have valid actigraphy data for year 1 but valid data for year 2. Consequently, we analyzed the first valid actigraphy assessment period available for each participant. This study was approved by the UCSD Institutional Review Board and was carried out in accordance with the Declaration of Helsinki. All participants completed informed consent prior to study involvement. Data were gathered from wrist-worn Actisleep-BT device which measures raw acceleration data in gravitational units using a tri-axial accelerometer sampling at 30Hz. Over continuous 24-hour wear periods, this device can be used to monitor movement allowing for estimation of sleep/wake patterns similar to methods described by Ancoli-Israel et al. . In-bed and out-of-bed times were set based upon both actigraphy data and information from morning surveys on a cell phone survey. If participant sleep records were missing entirely, sleep onset and awake time were manually determined by a specially trained research assistant using the detection methods outlined by Full, et al. . TST, WASO, PS, and NA were computed based upon these in/out bed intervals. To construct our composite score, in the absence of publicly-available norms, we chose to normalize BD subjects’ sleep measures based upon the HCs as a normative sample. We first computed means and standard deviations of actigraphy sleep measures of TST, WASO, PS, and NA in the HC group. Based upon this, we created standardized z-scores in the BD group by subtracting each BD individual’s sleep measure from the means of measures in HCs, and dividing by the SDs in the HC group. We multiplied WASO and NA by −1 to keep variables in the same direction . We then computed the mean of z-scores across all measures to create our composite.After computing the individual sleep statistics and composite z-scores, we sought to determine whether the sleep was of overall worse quality in BDs as compared to HCs by testing if the z-scores were different from 0 .

We then assessed the correlation of the composite with demographic and clinical variables using Pearson correlations and t-tests . Finally, for variables significantly correlated with the composite sleep z-score, we conducted LASSO regression to identify the individual sleep statistics most contributing to this correlation. The demographic and clinical variables served as outcomes and predictors were the z-scores for TST, WASO , PS, and NA . All analyses were completed in Stata SE version 15 . Past research on sleep in BD has focused on individual actigraphy measures , and there is a need for a global measure of sleep derived from actigraphy. In this study, we explored whether a composite score would relate to expected clinical, cognitive, and biological factors and whether including multiple sleep variables was important to the observed relationships. While the composite score was related to variables in BD including gender, employment, medication load, and mania symptoms, these associations were for the most part driven by only one of the individual sleep metrics with the exception of medication load for which TST and NA jointly contributed. Our results suggest a composite score does not yield gains in predictive power over individual sleep metrics. In studies of BD, it may be more appropriate to choose metrics to examine based on theory rather than summarizing multiple sleep metrics together. Our finding that a composite score is not as informative as examining individual sleep metrics alone may be explained by a number of factors. We focused on the averaged selected sleep indices measured over a two-week period, and it may be that examining changes in patterns of sleep over time, rather than means of sleep indices, could better capture more variability and thus compute a more nuanced composite score. Our goal, however, was to identify an intuitive approach for summarizing poor sleep in clinical practice, and thus we only focused on commonly used and often readily available sleep metrics. It is possible, however, that incorporating measures of circadian patterns would better support a composite measure approach. Additionally, population-based norms of actigraphic sleep indices do not currently exist, and thus we chose to use our healthy control sample as the norms for construction of z-scores. It is important that more research be conducted to generate norms for the purpose of computing composite measures as we sought to do in this study. While a sleep composite may not provide additional utility beyond that of individual measures themselves, there was substantial variability in z-scores indicating there are persons with BD who may have worse or better sleep as compared to HCs by as much as one standard deviation. This highlights the need for future research to identify cut-points based on z-scores identifying the worse sleepers. Similarly, it may be important to examine how daily fluctuations in sleep could be incorporated into a composite score. In BD, sleep often varies night-to-night and it could be that these nightly fluctuations are representative of poor sleep rather than simply average sleep measures across nights. While clinical relationships to the composite z-score were generally driven by only one sleep measure, we found correlations with clinical variables pertinent to BD confirming previous literature.