Multi-dimensional sleep and mortality: The Multi-Ethnic Study of Atherosclerosis

Abstract

Study Objectives

Multiple sleep characteristics are informative of health, sleep characteristics cluster, and sleep health can be described as a composite of positive sleep attributes. We assessed the association between a sleep score reflecting multiple sleep dimensions, and mortality. We tested the hypothesis that more favorable sleep (higher sleep scores) is associated with lower mortality.

Methods

The Multi-Ethnic Study of Atherosclerosis (MESA) is a racially and ethnically-diverse multi-site, prospective cohort study of US adults. Sleep was measured using unattended polysomnography, 7-day wrist actigraphy, and validated questionnaires (2010–2013). 1726 participants were followed for a median of 6.9 years (Q1–Q3, 6.4–7.4 years) until death (171 deaths) or last contact. Survival models were used to estimate the association between the exposure of sleep scores and the outcome of all-cause mortality, adjusting for socio-demographics, lifestyle, and medical comorbidities; follow-up analyses examined associations between individual metrics and mortality. The exposure, a sleep score, was constructed by an empirically-based Principal Components Analysis on 13 sleep metrics, selected a priori.

Results

After adjusting for multiple confounders, a 1 standard deviation (sd) higher sleep score was associated with 25% lower hazard of mortality (Hazard Ratio [HR]: 0.75; 95% Confidence interval: [0.65, 0.87]). The largest drivers of this association were: night-to-night sleep regularity, total sleep time, and the Apnea-Hypopnea Index.

Conclusion

More favorable sleep across multiple characteristics, operationalized by a sleep score, is associated with lower risk of death in a diverse US cohort of adults. Results suggest that interventions that address multiple dimensions may provide novel approaches for improving health.

Graphical Abstract

Statement of Significance.

Individuals with more favorable sleep across multiple sleep metrics had lower mortality rates. Because mortality associations implicated multiple sleep dimensions, interventions that simultaneously target multiple sleep characteristics may provide novel approaches for improving health.

Introduction

Multiple sleep characteristics may unfavorably influence health through distinct and complementary pathways, such as activation of inflammatory pathways, intermittent hypoxia, vascular remodeling, energy balance, and oxidative stress, adversely impacting major health outcomes and longevity [1–27]. Short sleep duration and sleep disordered breathing are associated with all-cause and cardiovascular mortality [2, 13, 28–33]. In addition, sleep that is disturbed [9, 34], irregular in timing or duration [35, 36], of poor quality [5], lower efficiency or longer sleep latency [4, 10], or characterized by less rapid eye movement sleep [4, 37], is associated with multiple adverse health outcomes. Thus, while short sleep duration and sleep disordered breathing are recognized public health problems and risk factors for chronic diseases [38], other dimensions of sleep such as irregularity are also emerging risk factors [35, 36, 39–41].

Although many epidemiological investigations of sleep and health outcomes focus on single dimensions of sleep, sleep characteristics cluster and individuals are influenced by multiple aspects of sleep health [42, 43]. Thus, interventions that simultaneously address several sleep risk factors may provide novel approaches to sleep and health promotion. Moreover, a multi-dimensional approach can provide a holistic “birds eye view” of sleep health outcomes relationships, provide context for interpretation of other sleep metrics, and identify sleep phenotypes amenable to intervention. Prior research has identified composite sleep phenotypes with increased risk of disease and/or mortality, such as “low sleep maintenance with late/variable timing” and possibly “insomnia with objective short sleep duration” [1, 11, 44–47]. Other research has created sleep scores that assign higher or lower values in overall sleep health to participants, demonstrating that those with more favorable sleep health (higher scores) tend to have more favorable mental and physical health, greater physical functioning, and lesser cardiometabolic morbidity, with composite metrics modestly yet consistently tending to outperform singular metrics [48–57].

A conceptual framework of multi-dimensional sleep health is Ru SATED: sleep Regularity, Satisfaction, Alertness, Timing, Efficiency, Duration [42]. We recently expanded this framework to create composite sleep scores, combining data from polysomnography, actigraphy, and self-report to consider metrics included in Ru SATED plus those that describe sleep disorders that are common in the population, and showed that more favorable sleep across multiple dimensions was more common in non-Hispanic White adults as compared to racial and ethnic minority adults, suggesting the utility of this approach for understanding health disparities [58]. In this paper, we tested the hypothesis that higher sleep scores (more favorable sleep) are associated with lower risk of mortality in a multi-ethnic cohort. We used an empirically-derived composite sleep score by Principal Components Analysis of 13 sleep metrics selected a priori, and used the first Principal Component which explained the most variance in sleep metrics. We further assessed which specific sleep characteristics showed the stronger associations with mortality.

Methods

Study design

The Multi-Ethnic Study of Atherosclerosis (MESA) is a prospective cohort study that enrolled 6814 adults aged 45–84 years free of evident CVD at baseline (2000–2002) [59]. Participants of four background groups (White, Black, Hispanic, Chinese) were recruited from areas near six U.S. academic centers (Wake Forest: Forsyth County, NC; Columbia: Northern Manhattan and the Bronx, NY; Johns Hopkins: Baltimore City and Baltimore County, MD; University of Minnesota, St. Paul, MN; Northwestern: Chicago, IL; and UCLA: Los Angeles, CA) and followed longitudinally to assess health outcomes and vital status. Interim examinations were conducted approximately every 2–5 years to obtain updated risk factor data.

At Exam 5 (2010–2012), 4077 MESA participants were invited to participate the MESA-Sleep exam (2010–2013). Participants were excluded if they lived too far from a field site (n = 141, 3.6%) or reported using a positive airway pressure device, an oral appliance, or supplemental oxygen (n = 147, 3.6%). Of the remaining 3789 eligible participants, 2261 consenting participants were enrolled in the sleep study which included polysomnography, actigraphy, and sleep questionnaires [60]. Principal Components Analysis was run on study participants with complete sleep information (n = 1814). After exclusion of missingness, the final analytic sample was 1726 individuals. Written informed consent was obtained from the participants, and the study was approved by institutional review boards at each field center and sleep reading center.

Exposure: sleep scores

Sleep was assessed by single-night, unattended polysomnography (PSG; Somté PSG, Compumedics Ltd., Abbotsford, Victoria, Australia), 7-day wrist actigraphy (Actiwatch Spectrum; Philips Respironics, PA), and validated sleep questionnaires, including the Epworth Sleepiness Scale and the Women’s Health Initiative Insomnia Rating Scale [61–64]. Sleep data were scored at Brigham and Women’s Hospital Sleep Reading Center using standardized approaches as described [65]. The Apnea-Hypopnea Index (AHI) was calculated as the average rate of apneas and hypopneas (hypopneas defined by 3% oxygen desaturation and scored based on 30% reduction in amplitude of the breathing signals) per sleep hour [66, 67]. Actigraphy data were processed by the Actiware-Sleep v 5.59 software (Mini Mitter Co., Inc., Bend, OR). Objective rest intervals (from “lights off” to wake) were determined by scorers based on participant-actuated event marker data from the Actiwatch, activity patterns, a sleep diary, and light sensors.

For this analysis we included 13 sleep metrics that we incorporated previously into a conceptual model that extended Ru SATED—sleep Regularity, Satisfaction, Timing, Efficiency, Duration [42, 43]: (A) from wrist actigraphy: (1) average nocturnal sleep duration, (2) timing/midpoint irregularity, (3) sleep duration irregularity, (4) sleep maintenance efficiency, (5) Fragmentation Index [68], (6) sleep onset latency, and (7) average midpoint timing; (B) from polysomnography: (8) Wake after sleep onset (WASO), (9) AHI, (10) % N3 (slow wave sleep), and (11) % R (rapid eye movement sleep), and (C) from questionnaire: (12) sleep quality (Women’s Health Insomnia Rating Scale score), and (13) sleepiness/alertness (Epworth Sleepiness Scale score). Additional justification for these metrics included prevalent concerns in the aging MESA population such as elevated AHI levels and greater sleep fragmentation [43, 58].

Principal Components Analysis was conducted using 13 sleep metrics standardized to mean = 0, sd = 1 [43, 58]. The first Principal Component—the component explaining the largest proportion of variance (19.75%)—represented contributions from all metrics to greater [e.g. fragmentation, efficiency, duration, midpoint variability) and lesser degrees (latency, quality) Supplementary Table S1]. Principal Component 1 was standardized and used as the sleep score (higher is better).

Outcome: all-cause mortality

The outcome was all-cause mortality; adjudicated events to 2018 are considered. Vital status was determined through telephone calls made at 9–12 month intervals [69]. Adjudication of deaths included review of death certificates, medical records of hospitalizations, next of kin interviews, and through annual National Death Index searches. Additional endpoint information was obtained by clinic visits, call-ins, medical record abstraction, or obituaries.

Covariates

In primary analyses, we adjusted for socio-demographics and lifestyle factors. Socio-demographics included: age, sex (male, female), self-reported race or ethnicity (non-Hispanic White, Black, Chinese-American, or Hispanic), degree attainment (less than high school, high school or some college, college degree, or graduate degree), household income (tertiles), and marital status (married/living with a partner, other). Lifestyle factors included: moderate-vigorous physical activity (by MESA’s Typical Week Activity Survey [70], metabolic equivalents of task), smoking status (never, current, former, not known), and diet including alcohol consumption (Alternative Healthy Eating Index—10) [71, 72]. In a subsequent model, we also included Body Mass Index (BMI, weight in kg/height in m²); and prevalent cardiovascular disease (CVD).

Sensitivity models adjusted for factors that might be on intermediate pathways, were correlated with other measures, or collected at varying timepoints: depressive symptoms (Center for Epidemiologic Studies—Depression [73], minus “restless sleep” question), hypertension and diabetes at Exam 5, prevalent cancer, chronic lung disease, total medication use, and pack years of smoking; hours second-hand smoke exposure per week (measured at Exam 4; 2005–2007); health insurance status (Exam 3; 2004–2005).

Analytic strategy

We used multivariable Cox Proportional Hazard models to relate sleep scores to time to death, and to estimate hazard ratios (HR) and 95% confidence intervals (CI). We sequentially adjusted for blocks of covariates: socio-demographics (model 1), lifestyle factors (model 2), and body mass index and CVD (model 3). We conducted a likelihood ratio test and computed measures of explained variation for inclusion of sleep scores beyond baseline socio-demographic and lifestyle factors. We checked the proportional hazards assumption by testing association of the standardized Schoenfeld residuals with follow-up times. We assessed the linearity of mortality associations with sleep scores by visualizing associations using penalized smoothing splines (df = 3).

To address reverse causality, in a sensitivity analysis we excluded deaths that occurred within 1-year of the sleep exam (n = 1708; deaths = 153). We conducted further sensitivity analyses (see Covariates). We conducted a follow-up analysis on the full sample (n = 1726) that regressed time to death on continuous, individual, standardized sleep metrics entered singly into models, adjusting for socio-demographics and lifestyle, providing a qualitative assessment of component-level contributions. For this analysis, several sleep metrics such as AHI were reverse-coded to reflect “higher is better” for comparison across sleep characteristics. Analyses were conducted in R 3.6.3 and 4.1.2.

Results

Average participant age was 68 years, included 55% females and 639 White (37.0%), 472 African American (27.3%), 204 Chinese-American (11.8%), and 411 Hispanic American (23.8%) adults (Table 1). Participants tended to have at least high school degree or some college, most were former or never smokers, and had a BMI of 28.7 kg/m². As previously reported, MESA-Sleep participants are slightly healthier (e.g. 7.1% smokers vs. 8.4%), younger, and included more racial-ethnic minorities than Exam 5 participants without sleep data [65]. MESA-Sleep participants were comparable to Exam 5 participants in sex, BMI, diabetes, and prior myocardial infarction. During the median 6.9 [Q1 = 6.4, Q3 = 7.4] years of follow-up there were 171 deaths (68 in women, 103 in men).

Table 1.

Socio-demographic, lifestyle, and health descriptive statistics, Examination 5, The Multi-Ethnic Study of Atherosclerosis (n = 1726, 2010–2012), and deaths (2018)

	Overall	Alive	Deceased
n	1726	1555	171
Deaths	171 (9.9%)	–	–
Follow-up time (years)	6.92 [6.39, 7.40]	7.01 [6.51, 7.44]	4.40 [2.46, 5.81]
Age (years)	68.23 (9.03)	67.44 (8.64)	75.40 (9.37)
Female	928 (53.8%)	860 (55.3%)	68 (39.8%)
Race-Ethnicity
Non-Hispanic White	639 (37.0%)	572 (36.8%)	67 (39.2%)
Chinese-American	204 (11.8%)	195 (12.5%)	9 (5.3%)
Black	472 (27.3%)	413 (26.6%)	59 (34.5%)
Hispanic	411 (23.8%)	375 (24.1%)	36 (21.1%)
Education
Less than high school	241 (14.0%)	212 (13.6%)	29 (17.0%)
High school or some college	797 (46.2%)	712 (45.8%)	85 (49.7%)
College degree	340 (19.7%)	312 (20.1%)	28 (16.4%)
Graduate	348 (20.2%)	319 (20.5%)	29 (17.0%)
Income (tertile)
Lowest tertile	650 (37.7%)	562 (36.1%)	88 (51.5%)
Middle tertile	586 (34.0%)	533 (34.3%)	53 (31.0%)
Highest tertile	490 (28.4%)	460 (29.6%)	30 (17.5%)
Smoker status*
Never smoked	951 (55.1%)
Former smoker quit more than 1 year ago	653 (37.9%)
Current smoker	111 (6.4%)
Alternative Healthy Eating Index—10	55.32 (10.09)	55.35 (10.15)	55.11 (9.58)
Moderate-Vigorous Physical Activity (MET)	5497.41 (6383.59)	5614.70 (6534.83)	4430.86 (4675.42)
Body Mass Index (kg/m2)	28.67 (5.54)	28.68 (5.52)	28.61 (5.71)
Prevalent cardiovascular disease (any)	104 (6.0%)	86 (5.5)	18 (10.5)
Sleep health score (Principal Component 1)	0.01 (1.00)	0.05 (0.98)	−0.38 (1.09)

^*11 participants reported “Do not know” for smoking.

Figure 1 shows predicted survival probabilities for sleep scores, at +1 and −1 sd above and below the mean of the continuous sleep score: those with lower sleep scores (−1 sd) were predicted to have lower survival probability over time than those with higher sleep scores (+1 sd). Table 2 shows results of the Cox proportional hazards regression analysis. One standard deviation higher values in the PC1-based sleep score were associated with a 25% lower mortality risk (HR: 0.75 [0.65, 0.87]) after adjustment for socio-demographics and lifestyle factors. Figure 2 shows that the sleep score (PC1) has an approximately linear relationship with mortality. The test of non-linearity in the spline model was not significant (p = .99). Further adjustment for (1) BMI and CVD (Model 3), (2) exclusion of deaths within 1-year of the sleep exam (n = 1708, deaths = 18), (3) exclusion of those with prevalent CVD (n = 1470, deaths = 131), (4) adjustment for additional covariates did not substantively change estimates beyond analyses which adjusted for socio-demographics and lifestyle factors (Model 2).

Figure 1.

Survival probability estimates over time at 1 standard deviation (SD) above and below the mean (1 SD, −1 SD) (n = 1726).

Table 2.

Cox proportional hazard model results for mortality regressed on the Principal Components-based sleep score

	Sleep score (continuous, per 1 sd)
	Hazard Ratio [95% CI]
Model 1: Socio-demographics	0.75 [0.65, 0.87]
Model 2: Lifestyle factors	0.75 [0.65, 0.87]
Model 3: Body Mass Index, prevalent disease	0.75 [0.65, 0.88]

The Multi-Ethnic Study of Atherosclerosis (n total = 1726; n deaths = 171). Model 1 adjusted for Socio-demographics (age, sex, race-ethnicity, education, income, and marital status). Model 2 adjusted for variable in model 1 plus Lifestyle factors (smoking, diet, moderate-vigorous physical activity). Model 3 adjusted for variables in model 2 plus Body Mass Index and prevalent disease (cardiovascular disease, cancer, chronic obstructive pulmonary disease). Hazard ratios are per 1 standard deviation increase in the score. Sample size n = 1726, 171 deaths.

Figure 2.

Spline-modeled Principal Component 1 sleep scores. The Multi-Ethnic Study of Atherosclerosis (n = 1726).

Figure 3 shows the results of time to mortality regressed on individual, standardized sleep metrics, each entered separately in models adjusting for socio-demographics and lifestyle factors. These analyses suggested that the strongest metrics associated with mortality were total sleep time and sleep midpoint (timing) and duration variability (from actigraphy) and the AHI and %REM sleep (from polysomnography), with hazard ratios ranging from 0.79 to 0.86 per standard deviation change. A likelihood ratio test (χ² = 13.7, p < .001) and generalized R² suggested added value of the sleep score over and above individual metrics (Supplementary Table S3). An exploratory analysis (not shown) included the 2nd and 3rd Principal Components, which were not associated with mortality (p_PC2 = .20; p_PC3 = .18).

Figure 3.

Follow-up analyses regressing mortality on the sleep score PC1 and specific, standardized components of the sleep score (entered singly), adjusting for socio-demographics and lifestyle. The Multi-Ethnic Study of Atherosclerosis (n = 1726, deaths = 171).

Discussion

We found in a multi-ethnic, aging U.S. cohort that a composite sleep score derived using Principal Components Analysis that summarized systematic patterns across multiple sleep metrics (including those that capture sleep patterns and sleep disorders) was associated with all-cause mortality after adjustment for socio-demographics and lifestyle factors such as smoking status, diet, moderate-vigorous physical activity, as well as major medical comorbidities. Analyses suggested that the relationship between sleep health (as operationalized in this study) and mortality is approximately linear, with higher sleep scores tracking with lower mortality hazard. Findings were robust in several sensitivity analyses that excluded individuals who died within 1-year of the sleep exam, or were additionally adjusted for multiple comorbidities and health-related risk factors, highlighting the public health and clinical importance of sleep.

Individuals with low sleep scores compared to those with high sleep scores are expected to be burdened with not just one sleep risk factor but several risk factors simultaneously, for example, higher apnea severity during a sleep period of lesser duration and lower efficiency and timed irregularly from night-to-night. In contrast, individuals with high sleep scores tended toward regularly obtaining sufficient and efficient sleep concurrent with lesser apnea severity (Figure 1). This pattern of sleep dimension clustering and its association with mortality suggest value in considering “healthy sleep” in a similar manner as “healthy diet” is conceptualized as a public health target; that is, based on a pattern of multiple nutrients intake [74]. While the American Heart Association’s (AHA) highlighted the benefits of a multi-nutrient healthy diet, its recently updated guidelines (Life’s Essential 8) that now includes sleep health only identified sleep duration as a health target [74]. Our data suggest that a multi-component approach that incorporates information from multiple facets of sleep together better describes the beneficial aspects of sleep compared to individual dimensions, and therefore, multi-dimensional measures of sleep health may serve as better health targets than sleep duration alone.

Multi-dimensional sleep composite measures: considerations within an evolving area of research

It is important to recognize that while emerging data supports the multi-dimensional nature of sleep, there is not yet consensus on approaches to incorporate a multiplicity of mutually-informative sleep metrics within a systematic framework. Moreover, different sleep scores are developed for different goals and settings. For example, Ru SATED scales focus on domains readily characterized using questionnaires, focusing on mostly conceptually independent domains and does not explicitly consider common sleep disorders, such as sleep apnea [42, 43]. In contrast, we capitalized on the empirical non-independence (correlations) among a larger number of sleep metrics collected using a variety of tools (polysomnography, actigraphy, questionnaire), using an unsupervised learning approach, Principal Components Analysis, for theoretically-supported dimension reduction among diverse metrics to advance concepts and empirical research [43, 58].

Incorporation of sleep information from diverse sources has the advantage of integrating complementary information from device-based and self-reported data, consistent with the increasing recognition of the importance of considering both patient-reported and functional test data. This approach also allowed consideration of metrics relevant for disease diagnosis (e.g. AHI for sleep apnea) as well as for quantifying degree of circadian and sleep disruption. On the other hand, self-report data used alone are more readily scalable. Ongoing advances in wearable technologies will likely improve the ability to incorporate device-based physiological data into comprehensive sleep metrics for future general use in population health.

Beyond composite measures to analysis of specific sleep characteristics

A high sleep score can reflect various combinations of attributes. However, due to the high correlations among components, one would expect that improving certain features (e.g. reduced sleep disordered breathing severity) would improve other features (e.g. sleep fragmentation, sleepiness). Thus, focusing on inter-correlations and identifying component-level drivers of the composite-level statistical signal, may yield insights into intervention targets and expected improvement. After hypothesis tests were conducted on the composite PC1 sleep score, in follow-up analyses, we qualitatively explored which specific sleep metrics of the composite score were most strongly associated with mortality. Objectively-assessed night-to-night variability in sleep timing and duration emerged as novel mortality risk factors. Sleep irregularity has been associated with obesity, metabolic dysfunction, and incident cardiovascular events independent of sleep duration [35, 36, 40, 41]. Wallace et al. found associations between sleep regularity, conceptualized as variability in wake-time, and mortality in univariate, but not adjusted, analyses in a cohort of older men. Although there is building consensus that sleep regularity is a factor for health, cohort differences and differences in operationalizing regular sleep-wake patterns, potentially account for differences across studies [41, 75, 76].

Beyond regularity, shorter actigraphy-estimated sleep duration was associated with early mortality. Self-reported sleep has been reported to relate to mortality following a U-shaped curve, in which risk increased among both short and long sleepers. However, self-reported sleep may over-estimate sleep duration in comparison to actigraphy. Our analyses of actigraph-based sleep identified only a limited number of long sleepers, limiting inference on non-linear relationships [6, 77]. For sleep disordered breathing, results were consistent with prior literature which suggests that higher AHI level—reflecting more severe sleep apnea with its attendant sleep fragmentation, intermittent hypoxia, and episodic airway obstruction—increases risk for incident hypertension, diabetes, heart failure, and mortality [13, 15, 28, 78].

Polysomnography-assessed measures of sleep architecture—specifically WASO and %REM sleep—also were identified as suggestive markers or contributors to mortality. Higher WASO is a measure of sleep fragmentation: increased wakefulness during the sleep period has been observed to be a predictor of early mortality [4, 10]. Of sleep stages, REM sleep was found to associate with mortality risk. This finding is consistent with a recent multi-cohort study that reported that of multiple sleep measures, reduction of REM sleep (<15%) was predictive of early mortality [37]. In total, the current study supports the importance of sleep regularity, duration, efficiency, REM, and sleep apnea as inter-related targets for intervention.

Strengths and limitations

There are several limitations to consider in interpreting the results of these analyses. Although the observational design limits causal inference, the exclusion of early deaths attenuates the possibility of reverse causality. The modest number of deaths precluded inference on associations with specific causes of death, although future research may have more power to estimate the association between multi-dimensional sleep and cause-specific mortality.

Although we performed several sensitivity analyses to account for major comorbidities, residual confounding from unmeasured health status is possible. Analyses were based on data from a single cohort and further research should assess generalizability. In the current analyses, we used a sleep health index that we developed in several previous papers that examined racial-ethnic disparity and properties of multi-dimensional sleep [43, 58]. We chose representative measures of sleep continuity, fragmentation, depth, variability, latency, and sleep disordered breathing from actigraphy and PSG data. However, it is possible that an alternative, either an expanded or more parsimonious set of variables, may be equally or more useful for a given research project or clinical/public health setting. There is likely value in scores that: (1) leverage high dimensional data (for increased predictive ability); and those that (2) utilize a more parsimonious set of variables (e.g. capturing average as well as variability in sleep duration). Analysis of individual sleep dimensions did not identify questionnaire-based measures of sleep (such as quality) to predict mortality [5]. It is possible that alternative instruments or use of cutoffs for defining health would have yielded new associations. In addition, of 13 Principal Components, we used the first PC which had a health interpretation, explained the most variance, and which described coherent trends in individual sleep metrics. The patterns of loadings for the other PCs did not vary in a uniform direction, but still may contain phenotypes of interest. In exploratory analyses, however, the addition of these PCs to the Cox regression model did not appreciably change the study findings. While a single PC was used that was constructed from multiple dimensions of sleep health, further work is needed to fully understand the extent to which summary measures such as this represent singular or multiple latent constructs. Additional areas for development include incorporating data from multiple nights of sleep disordered breathing assessments (as may be possible with emerging wearable devices), and thus also capture variability in measures of sleep disordered breathing.

Conclusion

Higher sleep scores were associated with reduced mortality hazard in a diverse and aging cohort. Our work supports the potential utility of deriving summary measures of sleep health that focus beyond a single dimension (average duration), in a manner similar to the AHA’s approach for recommending optimal diet. The specific sleep characteristics of largest association (lowest hazard) were greater objective night-to-night sleep regularity (timing and duration), optimal sleep duration, and lower sleep apnea severity. These sleep characteristics are notable in that they are also targetable or modifiable. Consideration of health-relevant sleep dimensions may lead to novel sleep interventions that not only improve multiple aspects of sleep but may further improve health.

Supplementary material

Supplementary material is available at SLEEP online.

Disclosure Statement

Dr. Bertisch reports consulting fees from Eisai, Inc, Merck Sharp & Dohme Corp, Idorsia Pharmaceuticals, ResMed, and Optum Health. Dr. Wallace reports consulting fees from Noctem Health, Health Rhythms, and Sleep Number Bed. Dr. Redline reports consultant fees from Eisai, Inc, Apnimed Inc, and Jazz, Inc.