Multidimensional Sleep Health: Concepts, Advances, and Implications for Research and Intervention.

The new paradigm of multidimensional sleep health ("sleep health") offers both challenges and opportunities for sleep science. Buysse (2014) has described sleep health to be multidimensional, framed as positive attributes, operationalizable into composite measures of global sleep health, sensitive to upstream exposures, and consequential for downstream health. We highlight two paradigm-shifting effects of a multidimensional sleep health perspective. The first is the use of composite sleep metrics which i) enable quantification of population shifts in sleep health, ii) with possibly reduced measurement error, iii) greater statistical stability, and iv) reduced multiple-testing burdens. The second is that sleep dimensions do not occur in isolation, that is, they are commonly biologically or statistically dependent. These dependencies complicate hypothesis tests yet can be leveraged to inform scale construction, model interpretation, and inform targeted interventions. To illustrate these points, we i) extended Buysse's Ru SATED model; ii) constructed a conceptual model of sleep health; and iii) showed exemplar analyses from the Multi-Ethnic Study of Atherosclerosis (n=735). Our findings support that sleep health is a distinctively useful paradigm to facilitate interpretation of a multitude of sleep dimensions. Nonetheless, the field of sleep health is still undergoing rapid development and is currently limited by: i) a lack of evidence-based cut-offs for defining optimal sleep health; ii) longitudinal data to define utility for predicting health outcomes; and iii) methodological research to inform how to best combine multiple dimensions for robust and reproducible composites.

Architecture; and the Sleep Health Score (comprehensive indices across domains). Sensitivity 1 analyses assessed a parsimonious version of the Sleep Health Score (PC1) that eliminated 2 potentially redundant measures. Composite score internal reliability was assessed by alpha 3 Cronbach. Consistency in PC weights -both direction and magnitude -was evaluated across 4 each component. Trends among individual sleep health variables (e.g., duration, sleepiness) in 5 relation to SHS categories were reported. Global sleep health variations by racial/ethnic group 6 were assessed for potential utility for sleep health disparities research in the non-elderly. 7 Outcomes of composite sleep health scores (linear regression), and their dichotomized 8 components (modified Poisson regression (24)), were regressed on the exposure of race-ethnicity 9 (White=ref), with adjustment for age and sex. Analyses were conducted in R 3.6.3. 10 11
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

10
There is growing interest in considering sleep health as not merely the absence of a 11 disorder but as a summary of the positive attributes of healthy sleep. Similarly, there is 12 movement towards using a conceptual framework that articulates sleep as multidimensional, 13 potentially summarized as a composite measure (1). We further suggest the value in considering 14 the intrinsic correlations and interactions among sleep measurements that comprise global sleep 15 scores; the challenges in developing optimal composite indices; and the potential value in 16 extending sleep health scores to include additional dimensions that describe common sleep 17 disorders (especially in middle-aged and older populations), quantify sleep architecture, and 18 consider regularity in not only sleep timing but sleep duration. 19 In evaluating alternative ways to summarize sleep health in an aging multi-ethnic 20 community sample, we found evidence to support the utility of composite sleep health scores 21 based on the Ru SATED framework as well as scores extended to include measures from PSG 22 and actigraphy. In extending the Ru SATED framework with the additional consideration of 23 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. interpretable principal components with more comprehensive scores evincing higher internal 3 reliability than simpler scores. PCA on original and extended Ru SATED frameworks showed 4 that the individual indices of sleep health for each score aggregated in consistent dimensional 5 patterns. The face validity of the extended SHS as a composite was supported by the empirical 6 finding that it summarized broad systematic shifts (most-least favorable categories; 0-3 vs 9-12) 7 across multiple sleep health metrics, including shifts in the initial components of Ru SATED as 8 well as in duration regularity, AHI, and sleep architecture. Moreover, each composite score 9 varied with race/ethnicity, supporting its utility for describing and monitoring sleep health 10 disparities. Adding additional sleep information also suggested important patterns of 11 association. For instance, timing and duration irregularity showed sleep disparities, which might 12 suggest that assessing determinants or barriers to (ir)regular bed and wake times may be 13 beneficial in reducing racial-ethnic sleep disparities rather than a focus on regular bed or wake 14 times alone (25). Although timing and duration regularity are moderately correlated, only about 15 31% of the variance of one accounts for the other, suggesting these regularity metrics to be 16 patterned yet distinct phenomena. For example, rotating shift-workers may have consistent 17 duration, yet inconsistent timing. Therefore, in middle-age or older cohorts where sleep disorders 18 and circadian disruption are prevalent, there is value in extending global sleep health assessments 19 with data from PSG and actigraphy. Moreover, metrics of OSA, altered sleep architecture, and 20 sleep duration variability predict adverse health outcomes, cognitive decline and mortality, 21 underscoring the potential utility of these data for informing sleep health assessments (26)(27)(28). 22 Our analyses not only supported use of composite indices, but also concurrent 1 consideration of multiple individual dimensions. For example, we found that the individual 2 "drivers" of differences in sleep health composites differed across race/ethnic groups, consistent 3 with prior work in both adults and the elderly (14). For example, compared to Whites, scores in  we further discuss some of the challenges and opportunities presented by multidimensional sleep 10 health, and analysis of both composite and individual dimensions. 11 Statistical correlation among sleep variables and implications for a composite sleep 12 health approach 13 We demonstrated that many sleep dimensions (both within and across domains) are inter- 14 correlated, reflecting their intrinsic physiological inter-relationships and potential responsiveness 15 to common stressors. Imagine, for instance, that each dichotomized SHS dimension in Table 1 is 16 represented by a light in a circuit. In the absence of dependence patterns, sleep health dimensions 17 would function like lights that burn out in parallel circuits, where health in one dimension does 18 not inform the others. However, if there are dependencies, much like a complex circuit with both 19 serial and parallel components, there is increased likelihood of some lights burning out (% NR, 20 sleepiness, sleep fragmentation) if certain other lights burn out (AHI). For instance, in our data, 21 higher sleep irregularity (timing and duration) and less sleep duration tended to co-occur. If this 22 pattern indicates the effects of an underlying common cause or latent factor, including measures 23 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. of both irregularity and duration should better characterize this driver within the distribution of a 1 composite. 2 In scales such as the SHS, therefore, the distribution of the composite score may reflect 3 relationships among the individual items that are informative. However, it is challenging to not 4 only select which measures to include in composites, but to determine the optimal way to 5 aggregate and weight each item, and whether to use continuous metrics or cutoff values 6 (discussed later). In choosing sleep variables for a composite, interpretability and ease of data  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. For public health utility, however, a categorical basis using prior knowledge to define 1 optimal ranges for each sleep dimension may also be appropriate. Early work operationalized Ru 2 SATED domains by dichotomizing daily diary, questionnaire, or actigraphy measures and 3 summing them into an overall sleep score (21)(22)(23). While continuous sleep exposures or 4 outcomes offer greater statistical power, categorical assessmentsthose involving a cut-point - 5 provide three practical benefits (30). First, cut-points aid clinicians in making decisions. Second, 6 prevalence can be defined and then used for needs assessment: e.g., approximately one third of 7 Americans do not meet sleep quantity recommendations (31). Third, cut-points enable 8 quantification of prevalence trends over time and across groups. Additionally, cut-points may 9 help avoid issues of non-linearities in exposure-response relationships, and cut-points are useful 10 for setting goals for public health initiatives. Categorical assessments may be appropriate even if 11 there is no evidence of a latent, internal discrete structure (taxon) for that dimension. As Kessler 12 (2002) notes, "there appears to be no taxon for high blood pressure," and yet cut-points for blood 13 pressure based on external criteria such as risk of stroke guide clinical and public health 14 decisions (30). 15 Implications of the categorical approach for population sleep health are illustrated in 16 Table 1. At the dimension level, needs assessment becomes clearer. That WASO ≤20 minutes, 17 for instance, only has 6.8% favorability in MESA may suggest a calibration of the NSF's optimal 18 ranges through extra data collection or may indicate an unmet need in middle-aged and older 19 adults (or both). Both cases are of public health import because age-appropriate optimal sleep 20 ranges can provide a set of public health targets to be validated against external criteria, whereas 21 unmet needs in a community suggest a need for investigation and intervention. 22 23 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021.   (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021.  other dimensions of sleep health may be expected (e.g. sleepiness, satisfaction, duration 12 variability, timing variability, etc.). For dimensions more difficult to directly target (% NR), 13 interventions could focus on correlated dimensions that are amenable to intervention (e.g., 14 irregularity, AHI). Examining multiple dimensions also may be useful in evaluating the effects 15 of interventions such as zolpidem, which may positively influence features such as sleep latency 16 and WASO but negatively influence sleepiness/alertness and NR sleep. Interventions that affect 17 only one sleep dimension may be the exception rather than the rule. 18 Multidimensionality opens the possibility of 'indirect' or 'root-cause' targeting, such as 19 targeting OSA to improve multiple sleep health metrics such as alertness, quality, continuity, and 20 % NR on a population level. Indeed, the high prevalence of OSA, its underdiagnosis, and causal 21 effects on other sleep dimensions suggest OSA as a driver of poor sleep health (19,(33)(34)(35), and 22 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ; thus an attractive intervention target to improve multiple sleep health dimensions (36)(37)(38). learning support this conclusion (7, 41).

13
The first limitation concerns the dependencies among dimensions, which may be one of 14 the more interesting and useful features of sleep health. The source(s) of these dependencies, 15 however, remain poorly understood. It is possible that the correlations observed are particular to 16 the sample or attributable to confounding factors such as the socioecological context, age or 17 other factors. 18 A second limitation concerns evolving or uncertain definitions and criteria of sleep 19 health. Consensus measures to assess sleep health with objective sleep data are lacking. We (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ; https://doi.org/10.1101/2021.04.20.21255799 doi: medRxiv preprint might estimate weights for each dimension in relation to a health outcome, with the limitation 1 that weights may be particular to the population and outcome. 2 The composites used self-reported and objective sleep indices. Even for the simple Ru 3 SATED score, which we derived by using objective data to define timing, duration and 4 efficiency, and from questionnaire data on quality and sleepiness/alertness, the PCA showed that  Moreover, explicit guidance on population specific thresholds is limited. The core 12 definition of Ru SATED dimensions is evolving as well, with sleep regularity a recent addition. 13 A canon of sleep health parameters for evolving measures of sleep micro-architecture (e.g. 14 spindles, k-complexes) has yet to be established despite emerging data on the unique information 15 contained in these measures (27,46). 16 These limitations highlight that sleep health is undergoing conceptual advances and scale 17 development (1, 6), scale validation (5, 6), implementations of sleep health in cohort studies and 18 community samples (21,47), innovations in methodological approaches (4,7,41,48), optimal 19 range and threshold determination (18,22), and many practical choices on how best to analyze 20 multidimensional sleep data. We suggest that statistical dependencies are an additional factor to 21 contend with. As sleep health continues to develop as a field, we suggest two solutions that may 22 help sleep health's future development.
23 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  3 From multidimensional/symptom psychiatric epidemiology (30), it is instructive to 4 observe the outcome of the tension between epidemiological theories of 'distribution shifts vs 5 targeted intervention' and policy making (30). Although shifting distributions is a powerful 6 public health approach, the current policy paradigm favors targeted intervention, prevalence 7 estimates, reporting prevalence trends over time, etc. (31). Thus, identifying an initial threshold 8 target, with monitoring of impact of interventions on this target and related health outcomes, 9 provides an example of how sleep health research can be translated to policy. This approach is 10 similar to the long-standing practices of using blood pressure cut-offs as public health and 11 individual targets despite evolving data on specific thresholds that confer disease (30). Future 12 research to develop normative sleep ranges will further support the utility of this approach for 13 policy makers and the public health. Logically implied are similar needs for developing and 14 validating normative or optimal ranges for composite sleep metrics. 15 Advantages to using composite scores include the reduction of multiple testing burden (as 16 compared to analysis of multiple individual dimensions) and the ability to detect the effects of 17 multiple small effects. An analogy can be drawn between sleep health and nutrition, in which 18 most nutrients appear to contribute in small ways to a larger composite effect (and sometimes 19 this composite effect better describes nutritional profiles than when individual nutrients are 20 examined) (2). Given their simplicity, composites also have ready public health application, 21 similar to promoting diets such as the Mediterranean diet rather than recommending specific 22 nutrient targets. (2). However, while summary scales may be selected a priori based on extant 23 All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ; https://doi.org/10.1101/2021.04.20.21255799 doi: medRxiv preprint empirical or conceptual work, prospective data are needed to demonstrate that they predict 1 clinically meaningful outcomes. 2 Methodological advances in sleep health composites are anticipated with advances in 3 sleep knowledge (and vice versa), perhaps leveraging increasingly larger datasets (49). 4 Nonetheless, there is a need to recognize that the components of global sleep health scores will 5 likely vary according to the questions at hand (e.g., interest in a pediatric versus aging  Another tension appears to depend on sleep health's complexity and wide-ranging 11 importance. Because sleep health is multidimensional, whose dimensions may be variably 12 more/less responsive to a wide set of individual and social factors, and whose different 13 components may be more/less consequential for different outcomes, a single, canonical   (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ; https://doi.org/10.1101/2021.04.20.21255799 doi: medRxiv preprint sleep-wake patterns. (53) As these data are collected on larger numbers of individuals, it will be 1 important to consider whether they provide novel metrics of sleep timing, variability and others 2 features not apparent in shorter-term measurements.  14 Longitudinal data may be particularly useful to clarify specific sleep drivers and levers. 15 We suggest sleep regularity (in timing and duration) as candidates for further longitudinal 16 research due to their i) potentially modifiable nature (as reflected in sleep hygiene 17 recommendations for consistent bed and wake times), ii) consistent correlation with many other 18 sleep dimensions (Figure 3), and iii) ability to forecast incident metabolic dysfunction and 19 cardiovascular events (12,13). If regularity as a candidate driver of sleep health is supported by 20 longitudinal studies, then targeted interventions can be tested, including the impact on sleep 21 health composites and clinical endpoints. (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. In summary, comprehensive and repeated sleep assessment may inform the etiology and 1 'natural history' of multidimensional sleep health, identify candidate drivers and phenotypes, and 2 help establish normative sleep health ranges for scientific knowledge and as targets for public 3 health. To create robust sleep health scores, additional research showing their value in predicting 4 health outcomes is needed. 5 Incorporate qualitative data 6 Finally, qualitative research complements other research methods, providing information, of racially-ethnically diverse low-income adults, commonly stated barriers to "good sleep" were 10 work (or multiple) work schedules, consuming large quantities of soda in the evening, reluctance 11 to stop using personal electronics at bed-time, child-care, and financial worries (25). Further  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ;    (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. b: These dimensions have optimal ranges in which overexpression of this dimension is considered problematic.
All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint this version posted April 23, 2021. ;    (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.    All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Multiple related measures allowed (see discussion of composite scores) All rights reserved. No reuse allowed without permission.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.