Trajectories of daytime sleepiness and their associations with dementia incidence

Abstract

Several studies have associated daytime sleepiness with dementia risk, but it is unknown whether longstanding and emerging daytime sleepiness equally signal dementia risk, and whether other health factors explain these associations. In a prospective, population-based epidemiologic study, we (1) assessed associations of daytime sleepiness trajectories over ten years with dementia incidence, and (2) examined whether selected health characteristics attenuated these associations. Using latent group-based trajectory analysis we categorized participants into three groups: (1) no daytime sleepiness (n=959, 49.2%); (2) emerging daytime sleepiness (n=342, 17.5%); and (3) persistent daytime sleepiness (n=650, 33.3%). Compared with no daytime sleepiness, emerging and persistent daytime sleepiness were similarly associated with greater incident dementia risk (respective hazard ratios (95% confidence intervals) were 2.2 (1.3, 3.5) and 1.9 (1.2, 3.1). Baseline blood pressure, body mass index, chronic disease diagnoses, and depression symptoms did not attenuate these associations. In contrast, dependence in instrumental activities of daily living attenuated the daytime sleepiness-dementia association by approximately 17-21%. These findings suggest that persistent and emerging daytime sleepiness may signal dementia risk. However, the underlying mechanisms remain unclear. Further studies should investigate whether and how pathways to sleepiness, functional impairment, and dementia pathophysiology inter-relate and manifest together over time.

Several past studies have demonstrated that cross-sectionally measured excessive daytime sleepiness is associated with future dementia risk (Foley et al., 2001; Merlino et al., 2010) and cognitive decline (Keage et al., 2012; Tsapanou et al., 2016). But little is known regarding the role of daytime sleepiness exposure duration, i.e., it is unclear whether effects of daytime sleepiness on dementia risk depend on if sleepiness recently emerged or has persisted over a long period. Dementia, at least that due to Alzheimer disease (AD), is the clinical manifestation of brain pathology that begins many years earlier, i.e. it has long latent and preclinical phases (Villemagne et al., 2013). If clinical dementia onset is associated with recently emerging daytime sleepiness, this would suggest that sleepiness is a marker of impending dementia. If, on the other hand, dementia onset is associated with only chronic longstanding daytime sleepiness, that would suggest that sleepiness is a potential independent risk factor; i.e., with traits, factors that cause or perpetuate sleepiness, or longstanding exposure to sleepiness itself, increasing dementia risk.

In addition, while past studies have demonstrated that sleepiness-cognition associations are independent of other health characteristics (Foley et al., 2001; Keage et al., 2012; Merlino et al., 2010; Tsapanou et al., 2016), we are unaware of prior reports regarding the extent to which specific health factors attenuate this association. Chronic diseases, for example sleep apnea and depression, are often associated with daytime sleepiness and may themselves increase dementia risk (Diniz et al., 2013; Leng et al., 2017). Daytime sleepiness could also lead to lower levels of physical activity and social engagement, which both may increase dementia risk (Penninkilampi et al., 2018; Stephen et al., 2017). If specific health characteristics explain a portion of the sleepiness-dementia association, this may shed light on the mechanisms that underlie the relationship between sleepiness and subsequent dementia.

In the current report, we first identify distinct trajectories of daytime sleepiness occurring over ten years in a large longitudinal population-based study of dementia risk. Next, we examine whether the identified trajectories were associated with dementia risk, and whether any specific health characteristics attenuated the observed sleepiness-dementia associations.

Methods

Sample:

Details regarding the Monongahela – Youghiogheny Heath Aging Team (MYHAT) study have been published previously (Ganguli et al., 2009). Briefly, 2036 participants were recruited using age-stratified random sampling based on voter registration lists; 1982 of them underwent detailed assessment and were re-assessed annually. Participants were required to be at least 65 years of age. The study was conducted in a small-town area near the Monongahela and Youghiogheny rivers south of the city of Pittsburgh, Pennsylvania. Almost all 96% (n=1951) of these participants had adequate data for this analysis and were included in the group-based trajectory analysis. Eight participants were excluded from the survival analysis because they already had dementia at study baseline. Written informed consent was obtained from all participants, and the study was approved by the University of Pittsburgh Institutional Review Board under protocol number STUDY19040058.

Primary predictor variable:

Excessive daytime sleepiness was assessed by self-report in response to the following question: “Do you ever fall asleep while actively doing something during the day?” Responses were coded never/rarely, sometimes, or usually. For the current analyses, responses were dichotomized as never/rarely vs. sometimes or usually. This decision was made because: (1) very few people responded “usually” (2.3% at baseline); and (2) never or rarely falling asleep while actively doing something duration the day indicates a low likelihood that sleepiness regularly impacts on daily function, whereas reporting this happening sometimes or usually indicates sleepiness that more frequently impacts daytime activities.

Covariates:

We included baseline covariates that, conceptually, could either confound or mediate the relationship between daytime sleepiness and dementia risk. Cigarette smoking, alcohol consumption, and exercise were assessed via self-report. Social engagement was measured by asking: “Do you belong to any organizations such as churches, lodges, societies, volunteer groups, etc.?” Responses of “yes” were coded to indicate the presence of social engagement. Depression symptoms were measured using the modified Center for Epidemiologic Studies Depression Scale (mCES-D) dichotomized at 3 symptoms to indicate the presence of at least mild depression symptoms (Ganguli et al., 2009). Overall health was captured based on total number of prescription medications being taken regularly at the time of assessment, dichotomized as less than 3 vs. 3 or more prescription medications. Dependence in instrumental activities of daily living (IADL) impairment was defined as needing help with at least one of the items on the OARS IADL scale (Fillenbaum, 1988). Self-reported physician diagnoses of the chronic diseases were recorded including cardiovascular disease, cerebrovascular disease, and diabetes.

Outcome variable:

The Clinical Dementia Rating (CDR ®) Dementia Staging Instrument (Morris, 1993) was used by trained interviews to rate participants after each annual study visit. The CDR is based on the participant’s independence in cognitively-driven everyday activities and functions, such that CDR=0 indicates normal functioning or independence, CDR=0.5 indicates mild cognitive impairment (MCI); CDR ratings of 1, 2, and 3 indicate mild, moderate, and severe dementia. Here, the outcome variable was incident dementia, recognized as the first visit at which the participant’s CDR was rated as ≥1. This is not a specific pathology-based neurocognitive disorder diagnosis such as Alzheimer’s or vascular dementia, but rather, a standard approach to rating the clinical dementia syndrome based on the presence of at least moderate functional impairment due to cognitive deficits.

Statistical analysis:

Since excessive sleepiness itself is associated with morbidity and mortality we recognized that our models examining whether sleepiness predicts dementia could be biased by non-random attrition from this aging cohort. Thus, we employed a joint modelling approach which simultaneously models two variables: daytime sleepiness and attrition due to death or illness. We used the function Jointlcmm from the R package lcmm to implement a joint latent-class mixed model accounting for non-random missingness. Our joint model includes two sub-models: (1) a linear mixed model of daytime sleepiness, and (2) a survival (time-to-event) model of dropout from the study due to death or illness, with Weibull baseline hazard. The linear mixed model explains daytime sleepiness, according to a trajectory of time, at the population level, and accounts for random effects on the intercept and time at the individual level. The optimal number of groups and the shape of trajectories (i.e., the longitudinal polynomials) were identified by using the Bayesian information criterion (BIC), which measures model fit improvements with a penalty for added complexity. Each participant is assigned to the latent-class in which he/she has the highest estimated posterior probability of membership. For descriptive purposes, we report covariate characteristics by the identified trajectory groups.

Next, we used the identified trajectories of daytime sleepiness as predictors of incident dementia, in Cox proportional hazard models with time-to-incident dementia as the outcome. The base model was adjusted for age, sex, and education. The proportional hazards assumption was examined via assessing the effects of time-covariates interaction. No adjusting covariates were significant at the 0.05 level, indicating that the proportional hazard assumption held. To identify covariates that attenuate the daytime sleepiness group-dementia associations, we fit models adding each covariate separately, and reported the adjusted sleepiness group hazard ratios (and percent attenuation from the base model).

Results

Loss to follow-up during the study:

Of the 1951 participants included in the analysis, 257 participants did not return after baseline and 458 died.

Sleepiness trajectory groups (Figure 1):

Figure 1. Latent-group trajectory plots for daytime sleepiness with 95% CI bands.

dots represent the predicted probability, lines represent observed probability and the shaded areas represent the 95% CI for observed probability.

Latent group-based trajectory modeling indicated that a four-group solution fit the data best (Supplemental Table 1 shows BIC for models with different numbers of groups). However, as two of these 4 groups appeared very similar, we selected a three-group solution combining the two similar groups. The groups were characterized by: (1) no daytime sleepiness (n=959, 49.2%); (2) emerging daytime sleepiness (n=342, 17.5%); (3) persistent daytime sleepiness (n=650, 33.3%). Participants in these groups differed in terms of several characteristics (Table 1) including medication use, cerebrovascular disease, and cardiovascular disease.

Table 1.

Baseline descriptive information by sleepiness trajectories

	Daytime Sleepiness
	No daytime sleepiness trajectory	Emerging daytime sleepiness	Persistent daytime sleepiness	p*
N	959	342	650
Age (mean (SD))	77.55 (7.48)	78.07 (7.17)	79.01 (7.44)	0.001
Female	607 (63.3)	216 (63.2)	369 (56.8)	0.021
Education				0.017
<High school	107 (11.2)	56 (16.4)	103 (15.8)
High school	434 (45.3)	147 (43.0)	300 (46.2)
>High school	418 (43.6)	139 (40.6)	247 (38.0)
Dementia	31 (3.2)	38 (11.1)	50 (7.7)	<0.001
Depression symptoms >3	48 (5.0)	22 (6.4)	51 (7.8)	0.067
Number of medications > 3	482 (56.8)	195 (61.3)	400 (65.5)	0.003
Instrumental activities of daily living dependences	118 (12.7)	55 (16.5)	146 (23.0)	<0.001
Smoker (ever)	500 (52.1)	184 (53.8)	347 (53.4)	0.821
Alcohol consumption				0.001
Never using alcohol in the past year	312 (32.5)	136 (39.8)	217 (33.4)
Consumed alcohol up to once per week	450 (46.9)	167 (48.8)	337 (51.8)
Consumed alcohol multiple times week	85 (8.9)	17 (5.0)	45 (6.9)
Daily drinkers	112 (11.7)	22 (6.4)	51 (7.8)
Cerebrovascular disease	110 (11.5)	58 (17.0)	94 (14.5)	0.024
Cardiovascular disease	364 (38.0)	137 (40.1)	289 (44.7)	0.027
Diabetes	191 (19.9)	81 (23.7)	153 (23.5)	0.145
Parkinson’s diseases	10 (1.0)	2 (0.6)	10 (1.5)	0.377
Systolic blood pressure ≥ 120	791 (82.9)	291 (85.6)	549 (85.1)	0.356
Diastolic blood pressure ≤ 70	342 (35.9)	127 (37.4)	296 (46.0)	<0.001
BMI				0.168
Normal (<25)	339 (35.3)	108 (31.6)	198 (30.5)
Overweight (25-29)	361 (37.6)	135 (39.5)	245 (37.7)
Obese (>=30)	259 (27.0)	99 (28.9)	207 (31.8)
Exercise	590 (61.5)	217 (63.5)	351 (54.0)	0.003
Physical activity	596 (62.3)	208 (60.8)	401 (61.7)	0.889
Social engagement	793 (82.8)	294 (86.0)	507 (78.0)	0.004
Sleep apnea	61 (6.4)	38 (11.1)	64 (9.9)	0.006

^*P-value for continuous variables are calculated from ANOVA; p-value for categorical variables are derived from chi-square test.

Percentage (number) are shown unless otherwise noted. Cardiovascular disease is a composite variable indicating the diagnosis of any of: congestive heart failure, irregular heartbeat, cardiac arrest or heart attack. Cerebrovascular disease is a composite variable which indicates the diagnosis of stroke or TIA. Note that all chronic diseases are based on self-report of a health-care professional’s diagnosis.

Dementia risk associated with sleepiness trajectories (Table 2 and Figure 2):

Table 2.

Attenuation of sleepiness-dementia association related to selected covariates

		Survival model for dementia			Adjusted HR (% attenuation)
		HR	p**	95% CI	Emerging	Persistent
Base model
Age		1.13	<0.001	(1.10, 1.17)	-	-
Education	HS vs. < HS	0.78	0.292	(0.49, 1.24)	-	-
	> HS vs. < HS	0.64	0.088	(0.38, 1.07)	-	-
Sex	Female vs. Male	0.87	0.451	(0.60, 1.26)	-	-
Daytime Sleepiness trajectory (vs. none)	Emerging	2.16	0.002	(1.34, 3.48)	-	-
	Persistent	1.94	0.004	(1.24, 3.05)	-	-
Additional models (each covariate was added to the base model separately)
Depression symptoms	mCES-D >3 vs. ≤3	2.27	0.006	(1.27, 4.06)	2.07 (4.19%)	1.89 (2.73%)
Number of medications	>3 vs. ≤3	0.92	0.651	(0.63, 1.34)	2.12 (1.64%)	1.91 (1.80%)
Instrumental activities of daily living dependence	Any vs. none	3.75	<0.001	(2.49, 5.66)	1.71 (20.91%)	1.62 (16.72%)
Systolic blood pressure	≥120 vs. <120	0.9	0.702	(0.53, 1.53)	2.18 (−0.91%)	1.96 (−0.95%)
Diastolic blood pressure	≤70 vs. >70	0.98	0.912	(0.68, 1.42)	2.17 (−0.53%)	1.96 (−0.92%)
Body mass index	25-29 vs. <25	0.64	0.040	(0.42, 0.98)	2.15 (0.57%)	1.99 (−2.20%)
	≥30 vs. <25	0.79	0.330	(0.49, 1.27)
Parkinson’s disease	Yes vs. no	0.98	0.983	(0.13, 7.19)	2.16 (0.06%)	1.94 (0.05%)
Smoking	Ever vs. never	0.91	0.627	(0.62, 1.34)	2.17 (−0.47%)	1.95 (−0.26%)
Drinking	Up to once per week vs. never	0.63	0.125	(0.43, 0.93)	2.03 (5.82%)	1.97 (−1.50%)
	Multiple times per week vs. never	0.15	0.111	(0.04, 0.63)
	Daily vs. never	0.43	0.187	(0.18, 1.01)
Cardiovascular disease*	Any vs. none	0.96	0.841	(0.67, 1.39)	2.16 (−0.12%)	1.95 (−0.41%)
Cerebrovascular disease*	Any vs. none	1.46	0.105	(0.92, 2.32)	2.08 (3.48%)	1.92 (1.07%)
Diabetes	Yes vs. no	1.1	0.676	(0.71, 1.69)	2.15 (0.28%)	1.94 (0.21%)
Exercise	Yes vs. no	0.61	0.008	(0.42, 0.88)	2.21 (−2.29%)	1.89 (3.08%)
Physical activity	Yes vs. no	0.57	0.002	(0.39, 0.81)	2.12 (1.94%)	1.95 (−0.16%)
Social engagement	Yes vs. no	0.47	<0.001	(0.31, 0.72)	2.16 (−0.14%)	1.86 (4.54%)
Sleep apnea*	Yes vs. no	1.08	0.22	(0.56, 2.07)	2.15 (0.43%)	1.95 (−0.16%)

Acronyms: mCESD – modified Center for Epidemiologic Studies Depression Scale; HS – High school.

^*Note that all chronic diseases are based on self-report of a health-care professional’s diagnosis.

^**p-values are derived from the Wald test from the Cox proportional hazard model. All rows in “additional models” represent separate models.

Figure 2. Dementia-free survival by daytime sleepiness trajectory groups.

lines indicate the percentage of each group that are dementia-free and the shaded areas represent the 95% CI for observed probability.

Compared with the group that had no daytime sleepiness, the groups that had emerging and persistent sleepiness both had approximately twice the risk of developing dementia over time (respective hazard ratios (HR) (95% confidence intervals (CIs)) were 2.2 (1.3, 3.5) and 1.9 (1.2, 3.1).

Attenuation related to baseline health characteristics:

The dementia risk associated with daytime sleepiness groups was not substantively attenuated by adjustment for any health characteristics, except IADL impairment (which attenuated the associations by 21% for emerging sleepiness and 17% for persistent sleepiness,

Discussion

Our population-based findings validate several past studies (Foley et al., 2001; Keage et al., 2012; Merlino et al., 2010; Tsapanou et al., 2016) that have linked daytime sleepiness with cognitive decline and dementia. We extended current knowledge by demonstrating that daytime sleepiness is associated with dementia risk regardless of whether it is longstanding or emerging in later life. In fact, associations of persistent and emerging daytime sleepiness had similarly sized effects on dementia risk (see similar patterns of dementia-free survival in Figure 2). These results indicate that both the emergence and longstanding presence of daytime sleepiness are not benign age-related changes, but rather, could provide a warning sign that dementia processes are underway.

None of the health characteristics examined explained the dementia risk associated with daytime sleepiness. Sleepiness apparently marks or contributes to dementia processes above and beyond conceptually related markers that were measured in this study (including depression and alcohol use). We did find that the presence of IADL impairment explained a portion of this association. While it is possible that daytime sleepiness might contribute to IADL impairment (e.g., sleepiness decreasing vigilance to complete IADLs), it is perhaps most plausible that IADL impairment and daytime sleepiness both reflect the same early, underlying, biological processes that link them with dementia.

Indeed, recent work led by Carvalho provides initial evidence that daytime sleepiness is associated with having less grey matter (D. Z. Carvalho et al., 2017) and more β-Amyloid accumulation over time (Diego Z. Carvalho et al., 2018) in the brain. Excessive daytime sleepiness is also very common in stroke survivors (Ding et al., 2016), suggesting that cerebrovascular disease may have a role linking sleepiness with dementia. Differences in these biomarkers of dementia pathophysiology, which were not measured in this study. may either lead to or result from daytime sleepiness (and related processes). However, a limitation of this epidemiologic research is that we do not have in-depth measures of all aspects of clinical and subclinical disease which could potentially confound or mediate associations between daytime sleepiness and dementia. For example, we lack measures such as amyloid deposition and cerebrovascular disease markers. Future studies to explore these relationships further should include appropriate neuroimaging measures.

In addition, our measures of medical conditions (such as sleep apnea) are based on participants reporting that they were given these diagnoses by a health care provider. This approach likely underestimates their prevalence, does not indicate disease severity, and thus may contribute to unmeasured confounding. Our measure of daytime sleepiness relied on a single question that was categorized to reflect the likely presence of fairly frequent daytime sleepiness. This daytime sleepiness measure is crude compared with scales that assess sleepiness severity or objectively measure. As a result, our findings only pertain to the statistical effect of fairly frequently self-reporting daytime sleepiness. Future research is needed to better characterize whether finer grained frequency measures, sleepiness severity, or objective indices of sleep proclivity or wakefulness more strongly or specifically predict dementia risk.

In conclusion, our findings extend those of several other groups (Foley et al., 2001; Keage et al., 2012; Merlino et al., 2010; Tsapanou et al., 2016) by adding the finding that both longstanding and recently emerging daytime sleepiness may signal dementia risk. While future mechanistic studies are pending, clinicians who care for older adults with excessive sleepiness should attempt to investigate and address its causes, and be alert to the possibility of an incipient dementia process.