Sleep Disorders and Cognitive Aging Among Cognitively Impaired Versus Unimpaired Older Adults

Abstract

Background and Objectives

Sleep disorders often predict or co-occur with cognitive decline. Yet, little is known about how the relationship unfolds among older adults at risk for cognitive decline. To examine the associations of sleep disorders with cognitive decline in older adults with unimpaired cognition or impaired cognition (mild cognitive impairment and dementia).

Research Design and Methods

A total of 5,822 participants (M_age = 70) of the National Alzheimer’s Coordinating Center database with unimpaired or impaired cognition were followed for 3 subsequent waves. Four types of clinician-diagnosed sleep disorders were reported: sleep apnea, hyposomnia/insomnia, REM sleep behavior disorder, or “other.” Cognition over time was measured by the Montreal Cognitive Assessment (MoCA) or an estimate of general cognitive ability (GCA) derived from scores based on 12 neuropsychological tests. Growth curve models were estimated adjusting for covariates.

Results

In participants with impaired cognition, baseline sleep apnea was related to better baseline MoCA performance (b = 0.65, 95% confidence interval [95% CI] = [0.07, 1.23]) and less decline in GCA over time (b = 0.06, 95% CI = [0.001, 0.12]). Baseline insomnia was related to better baseline MoCA (b = 1.54, 95% CI = [0.88, 2.21]) and less decline in MoCA over time (b = 0.56, 95% CI = [0.20, 0.92]). Furthermore, having more sleep disorders (across the 4 types) at baseline predicted better baseline MoCA and GCA, and less decline in MoCA and GCA over time. These results were only found in those with impaired cognition and generally consistent when using self-reported symptoms of sleep apnea or insomnia.

Discussion and Implications

Participants with sleep disorder diagnoses may have better access to healthcare, which may help maintain cognition through improved sleep.

Alzheimer’s disease, Insomnia, Sleep apnea, Social determinants of health

Age-related cognitive decline involves many risk factors, such as obesity, diabetes, hypertension, smoking, depression, and physical inactivity (Barnes et al., 2011). Compared to these well-known risk factors, however, the effects of sleep disorders on cognitive decline have been still less clearly understood (Baumgart et al., 2015), despite the prevalence of sleep disorders increasing with age (de Almondes et al., 2016; Yaffe et al., 2014). Although age-related decline in sleep is common (Breheny et al., 2023), sleep abnormalities in later life, such as daytime somnolence, trouble staying awake, and longer sleep onset latency, are associated with increased incidence of cognitive impairment and greater cognitive decline (Derby et al., 2016). Sleep apnea and insomnia are prevalent in older adults and these sleep disorders often predict or co-occur with cognitive decline (Rozzini et al., 2018). Further, potential impacts of sleep disorders on cognitive decline and vice versa may differ between individuals with unimpaired cognition or impaired cognition including mild cognitive impairment (MCI) and dementia. Yet, empirical evidence regarding complex relationships between sleep disorders and cognitive decline among older adults with different cognitive conditions is sorely lacking.

The social determinants of health (SDoH) and social-ecological model of sleep and health perspectives conceptualize how sleep disorders and cognitive decline may be related to each other (Grandner, 2019; Majoka & Schimming, 2021). The main tenet of these perspectives is that one’s sleep and cognitive health are affected by multilevel factors that the individual is embedded in, including genetics and behaviors (individual-level), socioeconomic status and neighborhood (social-level), and healthcare access and quality (societal-level). Although the influence of multilevel factors on sleep and cognition has not yet been explicitly tested, existing evidence shows that sleep and cognition may share common risks. Researchers have studied individual sleep disorders and report significant associations with cognitive decline or impairment. For example, the association between obstructive sleep apnea (OSA) and cognitive decline has been well documented (Incalzi et al., 2004). Insomnia and sleep disordered breathing (SDB) are common in patients with dementia, and each of these may contribute to further cognitive impairment (McCarter et al., 2016; Shi et al., 2018). REM sleep behavior disorder (RBD) is also associated with cognitive decline (Ferman et al., 1999) and co-occurs with dementia with Lewy bodies (DLB; Chan et al., 2018). Yet, conflicting results also exist. For example, those with clinician-identified insomnia have better cognition, larger hippocampus volume, and lower amyloid beta (Aβ) than individuals without insomnia, whereas clinician-identified nighttime behaviors are associated with worse cognition, smaller hippocampus volume, and higher Aβ (Zawar et al., 2022). Moreover, the levels of Aβ among participants with sleep apnea tend to be lower than among participants without sleep apnea (Hegde et al., 2020). These results are intriguing and call for more research examining a wide range of sleep disorders to better understand their influence on cognitive decline.

Having one sleep disorder tends to predict having another sleep disorder, especially in later adulthood (Suzuki et al., 2017). Comorbid insomnia and other sleep disturbances are common in patients with neurodegenerative disorders, and sleep disorders seem to be different in each form of dementia due to different brain pathologies (Cipriani et al., 2015). A greater functional impairment has been found in older adults with insomnia co-occurring with SDB compared to those with neither condition (Gooneratne et al., 2006). Guarnieri et al. (2012) reported co-occurrence of multiple sleep disorders in those with MCI and dementia. They found that insomnia and excessive daytime sleepiness co-occurred frequently in persons with MCI or dementia; restless legs syndrome (RLS) was always associated with other sleep disorders; and SDB co-occurred in all study participants with RLS. In addition, circadian rhythm dysfunction and SDB are prevalent in persons with Alzheimer’s disease (AD) dementia; hypersomnia, RLS, and RBD are common in those with DLB; and SDB is often found in those with vascular dementia (Yaffe et al., 2014). Also, the link between RBD and Parkinson’s disease has been reported (Masaoka & Phillips, 2016). Furthermore, SDB is common in patients with frontotemporal dementia (McCarter et al., 2016). Together, these studies suggest the importance of assessing the frequency of co-occurring sleep disorders in older adults with cognitive decline.

Another important consideration is the longitudinal relationship between sleep disorders and cognition. Sleep disorders may contribute to cognitive decline, early MCI, and AD pathology (Boot et al., 2012; Irwin & Vitiello, 2019). Sleep architecture changes and sleep abnormalities may start years before cognitive decline (Peter-Derex et al., 2015). The progression of the disease in persons with AD also affects brain regions regulating circadian rhythm and sleep architecture (Minakawa et al., 2019). Poor quality of sleep leads to increased Aβ production and decreased Aβ clearance (Ettore et al., 2019; Holth et al., 2017; Ju et al., 2014), and tau pathology can induce sleep disturbances (Basta et al., 2019; Wang & Holtzman, 2020). Thus far, few studies have used longitudinal data on sleep disorders and cognition, limiting the ability to understand how the relationship unfolds over time (Guarnieri & Sorbi, 2015). The relationship may differ by cognitive state (i.e., unimpaired vs impaired cognition), yet there is little research comparing cognitive groups.

This current study examined the associations of multiple sleep disorders with cognitive change in older adults with unimpaired or impaired cognition (including MCI and dementia). Little is known about how the two most prevalent forms of sleep disorders (i.e., sleep apnea and insomnia) and the co-occurrence of multiple sleep disorders relate to cognitive change across degrees of cognitive impairment. Our general expectation was that greater cognitive decline over time would be observed among those with sleep apnea, insomnia, and a greater number of multiple sleep disorders (across apnea, insomnia, RBD, and other types) at baseline. We also expected that poorer cognitive performance at baseline would be associated with a greater increase in sleep disorders over time.

Method

Data

We analyzed data from the National Alzheimer’s Coordinating Center (NACC) (data freeze on 03/01/2021). The NACC accrues data from 39 U.S. Alzheimer’s Disease Centers (ADCs) funded by the National Institute on Aging (NIA). Data collection occurs approximately annually. The ADC Clinical Task Force has defined a standardized clinical data set, called the Uniform Data Set (UDS), which was used in the current study. We studied NACC data from participants with cognitive status ranging from unimpaired, to MCI, to dementia enrolled in an ADC since 2005. For detailed description of the NACC and UDS, see Besser et al. (2018) and Morris et al. (2006). Data collection was under the oversight of local IRBs associated with each ADC. Written informed consent was obtained from all participants and coparticipants (usually a close friend or family member).

Participant Selection

From the complete NACC data set (n = 43,517), we restricted our sample to participants who had a baseline visit beginning in March 2015, when UDS-3 was implemented. NACC began collecting data on sleep disorders at this time (n = 10,740). We further restricted our sample to participants with at least 2 visits over time so that we could study longitudinal associations between levels of and change in sleep disorders and cognition (n = 6,075). Finally, we excluded participants with a baseline diagnosis of impaired not MCI (n = 253), due to the small sample size and great phenotypic and etiologic variability of this diagnostic group. Our final sample included 5,822 participants diagnosed at baseline with having unimpaired cognition (n = 2,949) or impaired cognition (n = 2,873; including MCI [n = 1,480] and dementia [n = 1,393]). Although some participants had up to six waves of data available, we limited our longitudinal analyses to four waves due to extensive missingness at later times. Participants completed about three visits on average (SD = 0.97).

Measures

Sleep disorders

At each wave, ADC clinicians recorded whether participants reported they had the following clinician-diagnosed sleep disorders as present (=1 point each) or absent (=0): sleep apnea, hyposomnia or insomnia, REM sleep behavioral disorder (RBD), or “other” sleep disorders. Examples of “other” sleep disorders include restless leg syndrome, delayed sleep phase disorder, and narcolepsy among others. We used the binary variables of sleep apnea or hyposomnia/insomnia based on baseline identification by clinician. We also calculated the total number of sleep disorders participants had at each wave across the four types (range = 0 to 4). This variable was used to examine the effects of having multiple sleep disorders at baseline and increase in the number of sleep disorders over time on cognition. Using clinician-identified sleep disorders alone may not give us the whole picture, as sleep disorders are often underdiagnosed. To examine the discrepancy between % of symptoms versus % of diagnoses and whether and how the trajectory of cognition differs by the self-awareness of sleep problems, we supplemented our analyses with self-reported, current symptoms of sleep apnea and insomnia available based on the participant’s health history collected at the baseline assessment. Participants were classified as having sleep apnea or hyposomnia/insomnia if the disorder was recent/active (=1) and as not having sleep apnea or hyposomnia/insomnia if the disorder was absent or remote/inactive (=0).

Cognitive decline

To measure global cognitive function, we used the Montreal Cognitive Assessment (MoCA; Weintraub et al., 2017). MoCA was adjusted for level of education such that for individuals with 12 years or fewer of formal education, their score on the raw MoCA was increased by 1 point. Total scores on the MoCA in this sample ranged from 0 to 30. We also included data from the UDS-3 neuropsychological battery (Weintraub et al., 2017). Based on 12 specific tests used in prior studies (Kiselica et al., 2020), we created a general cognitive ability (GCA) score through a principal component analysis. See Supplementary Table 1 for the details of the 12 tests and how we calculated GCA scores. We followed an established process of calculating the GCA scores for Waves 1 through 4 based on prior research (Finkel et al., 2009).

Covariates

We included covariates that were sociodemographic or health-related. Sociodemographic covariates included baseline age (years), sex (male = 1, female = 0), race (two variables for [a] Black/African American or [b] Other race compared to White [reference]), ethnicity (Hispanic = 1, not Hispanic = 0), baseline marital/partnered status (married or partnered = 1; widowed, divorced, separated, or never married = 0), and educational attainment (years). Health-related covariates included baseline measures of BMI, which was dummy coded for individuals who were overweight/obese (BMI > 25) or underweight (BMI < 18.5), with healthy weight (18.5 ≤ BMI < 25) as the reference; smoking status (two variables indicating [a] current smokers or [b] former smokers compared to never smokers [reference]); use of antidepressant medications and/or anxiolytic, hypnotic, or sedative agents (yes = 1, no = 0); clinician-diagnosed depression and/or anxiety (yes = 1, no = 0); and the number of comorbid health conditions calculated based on the presence of clinician-assessed 15 medical conditions, 13 neurological conditions, 5 psychiatric conditions, and 5 other conditions (range = 0 to 13). See Supplementary Table 2 for specific conditions included. Full models included all these covariates, whereas reduced models included only covariates that were significantly and independently related to either sleep apnea and/or hyposomnia/insomnia at baseline (p < .05; Sauer et al., 2013). The list of covariates in reduced models included: baseline age, sex, Other race (vs White), ethnicity, educational attainment, overweight/obese, use of antidepressant medications and/or anxiolytic, hypnotic, or sedative agents, clinician-diagnosed depression and/or anxiety, and the number of comorbid health conditions.

Statistical Analysis

First, we conducted preliminary data screening to assess variables for normality, outliers, and heteroskedasticity. We then conducted descriptive statistics to describe the sample by baseline cognitive condition groups. Second, using growth curve models, we assessed whether cognition (MoCA and GCA) at baseline and over time differed by prevalent types of clinician-identified sleep disorders (i.e., apnea and insomnia) at baseline. Analyses were conducted separately for participants with unimpaired and impaired cognition, with follow-up analyses assessing associations in participants with MCI or dementia, separately. All study covariates were included in growth curve models with adjustments made to the intercept and slope. Time was modeled as number of waves from baseline, with the time coefficients set to 0, 1, 2, and 3, respectively, for the four waves. Random effects of the intercept and slope were included. Inclusion of quadratic slope effects did not significantly improve model fit, overall, and were therefore excluded from models.

Third, we assessed whether multiple (number of) clinician-identified sleep disorders were associated with cognition, using bivariate latent growth curve models (LGCMs). In these analyses, we examined both baseline and change in the number of diagnosed sleep disorders and linked them to baseline and change in cognition over time. The four timepoint measurements for sleep disorders and cognition were used to estimate latent intercept and slope variables for sleep disorders and cognition. We assessed whether the latent intercept or slope of sleep disorders was associated with the latent intercept and slope for cognition. Figure 1 illustrates the tested associations between sleep disorders and cognition. Similar to the previous models, we conducted separate analyses by baseline cognitive conditions. We ran models that adjusted both the X (sleep disorders) and Y (cognition) intercepts and the X (sleep disorders) and Y (cognition) slopes for covariates. To account for the skewed distribution in the number of sleep disorders variable and generate unbiased standard errors, we conducted bootstrapping with 5,000 bootstrapped samples. Bivariate LGCMs present standardized parameter estimates and the associations between latent intercepts and latent slopes are interpreted as correlations. Preliminary data screening, cleaning, variable creation, and growth curve models were completed using SAS v9.4. Growth curve models were conducted with PROC MIXED. Bivariate LGCMs were conducted in Mplus v8.3 (Muthén & Muthén, 2017). Significance was assessed at p < .05 (two-tailed).

Figure 1.

Model of tested relationships between number of sleep disorders and cognition. Observed sleep disorders represent a sum score of sleep apnea, insomnia, REM sleep behavior disorder, and a category for other sleep disorders. Cognition was measured with the Montreal cognitive assessment and a principal component analysis of neuropsychological test scores representing general cognitive ability. The model was run first by baseline diagnosis (unimpaired cognition and impaired cognition [including mild cognitive impairment [MCI] and dementia]). Sensitivity analyses were conducted separately for persons with MCI and dementia. The full model adjusted the latent intercepts and slopes of sleep disorders and cognition for covariates including baseline age, sex, race, baseline marital/partnered status, educational attainment, and baseline measures for all health-related covariates including body mass index, smoking status, use of antidepressant and/or anxiolytic medications, clinician-diagnosed depression and/or anxiety, and the number of comorbid health conditions. The reduced model controlled for baseline age, sex, Other race, ethnicity, years of education, overweight/obese, use of antidepressant and/or anxiolytic mediation, clinical diagnosis of depression and/or anxiety, and the number of comorbid health conditions. Paths for covariates not shown for clarity.

Results

Participant Characteristics

Table 1 contains baseline descriptors of participants by cognitive conditions. Compared to participants with MCI or dementia, those with unimpaired cognition were younger, more educated, women, non-White, and not married/partnered. There were no significant differences between groups regarding smoking status. Participants with dementia were less likely to be overweight or obese compared to participants with unimpaired cognition or MCI; yet there were no differences among groups in rates of underweight. Each group was different from each other in use of antidepressant or anxiolytic medications, with participants with dementia reporting the most frequent use, followed by participants with MCI, and the lowest use among participants with unimpaired cognition. Participants with unimpaired cognition were less likely to be diagnosed with depression or anxiety compared to participants with MCI and dementia. Participants with MCI reported more comorbid health conditions than participants with unimpaired cognition or dementia.

Table 1.

Baseline Descriptive Characteristics of Participants by Cognitive Conditions

Variable	Unimpaired cognition (n = 2,949)		MCI (n = 1,480)		Dementia (n = 1,393)		Differencec
	n	M (SD) or %	n	M (SD) or %	n	M (SD) or %
Age, years	2,949	68.57 (10.28)	1,480	71.94 (8.26)	1,393	69.61 (10.03)	*^+
Years of education	2,931	16.35 (2.71)	1,473	15.97 (3.11)	1,377	15.36 (3.22)	*^+
Sex							*^
Male	1,043	35.37%	724	48.92%	691	49.61%	—
Female	1,906	64.63%	756	51.08%	702	50.39%	—
Race							^+; ^d
White	2,256	77.82%	1,198	81.83%	1,214	88.87%	—
Black/African American	418	14.42%	178	12.16%	96	7.03%	—
Other race	225	7.76%	88	6.01%	56	4.10%	—
Hispanic							*+
No	2,697	92.08%	1,310	88.75%	1,288	92.60%	—
Yes	232	7.92%	166	11.25%	103	7.40%	—
Marital status							*^+e
Married/partnered	1,862	63.44%	1,026	69.70%	1,122	80.84%	—
Not married/partnered	1,073	36.56%	446	30.30%	266	19.16%	—
Smoking status							NSf
Never	1,773	60.66%	871	60.11%	861	63.45%	—
Former	1,039	35.55%	526	36.30%	437	32.20%	—
Current	111	3.80%	52	3.59%	59	4.35%	—
Body mass index	2,785	27.57 (5.55)	1,373	27.19 (4.94)		26.45 (5.17)	—
Underweight	28	1.01%	16	1.17%	21	1.70%	NS
Healthy weight	984	35.33%	453	32.99%	510	41.23%	—
Overweight/obese	1,773	63.66%	904	65.84%	706	57.07%	^+
Using antidepressants or anxiolytics							*^+
No	2,009	69.01%	855	57.97%	668	48.13%	—
Yes	902	30.99%	620	42.03%	720	51.87%	—
Diagnosed with depression or anxiety							*^
No	2,661	90.23%	1,166	78.78%	1,064	76.38%	—
Yes	288	9.77%	314	21.22%	329	23.62%	—
# Comorbid health conditionsa	2,949	2.37 (1.61)	1,480	2.95 (1.86)	1,393	2.49 (1.78)	*+
MoCA	2,830	26.43 (2.60)	1,366	22.55 (3.46)	1,224	16.05 (6.01)	*^+
GCAb	2,792	0.03 (0.98)	1,259	0.09 (0.97)	727	0.42 (0.94)	^+
Any sleep disorder	833	29.25%	553	38.16%	428	31.24%	*+
Number of sleep disorders in participants with sleep disorders		1.17 (0.40)		1.22 (0.46)		1.22 (0.47)	NS
Sleep apnea	390	13.76%	309	21.38%	243	17.75%	*^+
Hyposomnia/insomnia	450	15.49%	229	15.55%	139	10.01%	^+
RBD	67	2.34%	87	5.95%	104	7.51%	*^
Other sleep disorder	65	2.24%	50	3.41%	36	2.60%	NS

Notes: GCA = general cognitive ability; M = mean; MCI = mild cognitive impairment; MoCA = Montreal Cognitive Assessment; n = number of participants; NS = not significant; RBD = REM sleep behavior disorder; SD = standard deviation.

^aMeasured as a summary score of 38 clinician-assessed medical, neurological, psychiatric, and other conditions (range = 0–13).

^bMeasured through a principal component analysis from 12 neuropsychological test scores.

^cSignificant differences between diagnostic groups on variables assessed with one-way analyses of variance (ANOVAs) or Kruskal-Wallis tests where applicable. Significant differences at p < .05 between participants with (a) unimpaired cognition and MCI represented by *; (b) unimpaired cognition and dementia represented by ^; and (c) MCI and dementia represented by ⁺.

^dDifferences in race were tested among groups with the binary variables indicating Black/African American race and Other race, respectively (compared to White [reference]).

^eDifferences in marital/partnered status assessed with the variable indicating whether participants were (a) married or cohabitating with a partner or (b) had any other marital status.

^fDifferences in smoking status assessed with the binary variables indicating (a) former or (b) current smoking status compared to never smoking.

Scores on the MoCA were highest for those with normal cognition, lower for those with MCI, and lowest for those with dementia, with significant differences between each group. Although GCA was moderately correlated with MoCA (r = .37, p < .001), GCA scores differentiated only between those with dementia versus those with MCI or unimpaired cognition, with no difference between MCI and unimpaired cognition. About 31% of the sample had a clinician-identified sleep disorder, with higher prevalence of sleep disorders in those with MCI compared to other groups. Hyposomnia/insomnia (14%) and sleep apnea (16%) were the two most prevalent forms, and approximately 6% of the sample had more than one sleep disorder across waves. Among those with more than one sleep disorder at baseline (Table 2), the combination of sleep apnea and hyposomnia/insomnia was most common in those with unimpaired cognition (2.34%) and MCI (3.45%); in those with dementia, the combination of sleep apnea and RBD was most common (2.51%). The intraclass correlation indicates that 69% of the variance in the total number of sleep disorders was at the between-person level.

Table 2.

Baseline Prevalence of Sleep Disorder Diagnoses by Cognitive Conditions

	Unimpaired cognition (n = 2,949)	MCI (n = 1,480)	Dementia (n = 1,393)
Sleep apnea	287 (9.73%)	214 (14.46%)	174 (12.49%)
Insomnia	347 (11.77%)	156 (10.54%)	95 (6.82%)
RBDa	33 (1.12%)	45 (3.04%)	52 (3.73%)
Otherb	36 (1.22%)	26 (1.76%)	22 (1.58%)
Sleep apnea + insomnia	69 (2.34%)	51 (3.45%)	23 (1.65%)
Sleep apnea + RBDa	18 (0.61%)	26 (1.76%)	35 (2.51%)
Sleep apnea + otherb	7 (0.24%)	8 (0.54%)	4 (0.29%)
Insomnia + RBDa	10 (0.34%)	7 (0.47%)	8 (0.57%)
Insomnia + otherb	16 (0.54%)	6 (0.41%)	5 (0.36%)
RBDa + otherb	1 (0.03%)	4 (0.27%)	2 (0.14%)
Sleep apnea + insomnia + RBDa	4 (0.14%)	4 (0.27%)	5 (0.36%)
Sleep apnea + insomnia + otherb	4 (0.14%)	5 (0.34%)	1 (0.07%)
Sleep apnea + RBDa + otherb	1 (0.03%)	1 (0.07%)	0 (0%)
Insomnia + RBDa + otherb	0 (0%)	0 (0%)	1 (0.07%)
Sleep apnea + insomnia + RBDa + otherb	0 (0%)	0 (0%)	1 (0.07%)

Notes: MCI = mild cognitive impairment.

^aRBD = REM sleep behavior disorder.

^bOther types of sleep disorders include restless leg syndrome, delayed sleep phase disorder, and narcolepsy among others.

Associations Between Individual Sleep Disorders and Cognition by Cognitive Groups

We first established the trajectories of cognitive change in unconditional growth models separately for participants with unimpaired and impaired cognition. Participants with unimpaired cognition exhibited a nonsignificant increase in MoCA scores (b = 0.02, 95% CI = [−0.02, 0.07]) and a significant increase in GCA scores (b = 0.01, 95% CI = [0.002, 0.03]) across the four waves. Participants with impaired cognition showed significant declines in both MoCA (b = −1.53, 95% CI = [−1.64, −1.42]) and GCA (b = −0.19, 95% CI = [−0.22, −0.17]) scores across the four waves.

Table 3 presents results from growth curve models of cognition by clinician-identified sleep apnea or insomnia status at baseline in participants with unimpaired or impaired cognition. In participants with unimpaired cognition, MoCA and GCA did not differ by sleep apnea or insomnia. Adding in clinician-identified sleep disorders (apnea and insomnia) accounted for between 1% and 6% of the variance in interindividual differences in the intercept and between 0% and 16% of the variance in interindividual differences in the slope of cognition (measured with MoCA and GCA).

Table 3.

Growth Curve Model Results of Baseline Clinician-Diagnosed Sleep Apnea and Insomnia Related to Cognition for Participants With Unimpaired Cognition, Impaired Cognition, Mild Cognitive Impairment, and Dementia—Full Model

Variable	Unimpaired cognitiona			Impaired cognitionb			MCIc			Dementiad
	Estimate	LL	UL	Estimate	LL	UL	Estimate	LL	UL	Estimate	LL	UL
Clinician-diagnosed sleep apnea
Montreal cognitive assessment
Sleep apnea	0.10	−0.15	0.35	0.65	0.07	1.23	−0.08	−0.52	0.37	1.06	0.11	2.00
Time	1.04	0.53	1.56	−4.31	−5.59	−3.02	0.29	−1.20	1.79	−5.74	−7.70	−3.79
Sleep apnea × time	0.06	−0.08	0.21	0.15	−0.15	0.46	0.03	−0.28	0.33	0.10	−0.44	0.65
General cognitive ability
Sleep apnea	0.07	−0.02	0.16	0.06	−0.05	0.16	0.03	−0.10	0.16	0.11	−0.07	0.30
Time	0.41	0.27	0.54	0.02	−0.24	0.29	0.37	0.04	0.70	−0.55	−1.00	−0.11
Sleep apnea × time	0.02	−0.02	0.06	0.06	0.001	0.12	0.04	−0.03	0.10	0.11	−0.01	0.22
Clinician-diagnosed insomnia
Montreal cognitive assessment
Insomnia	−0.01	−0.25	0.22	1.54	0.88	2.21	0.45	−0.05	0.95	1.80	0.66	2.94
Time	1.09	0.57	1.60	−4.27	−5.54	−3.01	0.32	−1.14	1.78	−5.82	−7.75	−3.89
Insomnia × time	−0.002	−0.14	0.14	0.56	0.20	0.92	0.40	0.04	0.76	0.31	−0.34	0.97
General cognitive ability
Insomnia	0.02	−0.06	0.11	0.09	−0.03	0.22	0.09	−0.05	0.24	0.15	−0.07	0.37
Time	0.40	0.27	0.54	0.02	−0.24	0.29	0.37	0.05	0.70	−0.57	−1.02	−0.12
Insomnia × time	0.004	−0.03	0.04	0.03	−0.04	0.10	0.03	−0.05	0.11	0.02	−0.11	0.15

Notes: Sleep apnea and insomnia are dichotomized (1 = having clinician-identified disorder vs 0 = no).

Estimate = unstandardized parameter estimate; LL = lower limit of the 95% confidence interval; MCI = mild cognitive impairment; UL = upper limit of the 95% confidence interval.

Bolded values indicate significant effects at p < .05. Adjusted models controlled for baseline age, sex, race, ethnicity, baseline marital/partnered status, years of education, body mass index (dummy coded variables indicating underweight or overweight/obese compared to healthy weight), smoking status (dummy coded variables indicating current or former smoking status compared to never smoking), use of antidepressant and/or anxiolytic mediation, clinical diagnosis of depression and/or anxiety, and the number of comorbid health conditions. Both the intercept and slope effects were adjusted for all covariates.

^aapnea: n = 2,614; insomnia: n = 2,674.

^bapnea: n = 2,449; insomnia: n = 2,495.

^capnea: n = 1,294; insomnia: n = 1,321.

^dapnea: n = 1,155; insomnia: n = 1,174.

For participants with impaired cognition, MoCA performance at baseline was better (b = 0.65, 95% CI = [0.07, 1.23]) and the decline of GCA over time was less steep (b = 0.06, 95% CI = [0.001, 0.12]) for those with versus without sleep apnea. Similarly, MoCA performance at baseline was better (b = 1.54, 95% CI = [0.88, 2.21]) and the decline of MoCA over time was less steep (b = 0.56, 95% CI = [0.20, 0.92]) for those with versus without insomnia. Adding in clinician-identified sleep disorders (apnea and insomnia) accounted for between 0% and 2% of the variance in interindividual differences in the intercept and the slope of cognition (measured with MoCA and GCA). Follow-up analyses showed that participants with MCI and insomnia at baseline had less decline in MoCA over time (b = 0.40, 95% CI = [0.04, 0.76]). Participants with dementia and either sleep apnea (b = 1.06, 95% CI = [0.11, 2.00]) or insomnia (b = 1.80, 95% CI = [0.66, 2.94]) at baseline had better MoCA performance at baseline. Results from the reduced model were generally consistent with those from the full model (Supplementary Table 3).

Associations Between Multiple Sleep Disorders and Cognition by Cognitive Groups

Supplementary Table 4 presents fully adjusted bivariate LGCMs for participants with unimpaired cognition. Number of sleep disorders at baseline or change in the number of sleep disorders were neither related to cognition at baseline nor to change in cognition when measured with either MoCA or GCA. Results from the reduced model were consistent with those from the full model.

Figure 2 shows results for participants with cognitive impairment (either MCI or dementia). A greater number of sleep disorders at baseline was related to better MoCA (b = 0.09, 95% CI = [0.04, 0.15]) and better GCA (b = 0.06, 95% CI = [0.000, 0.11]) performance at baseline, and less decline in MoCA (b = 0.13, 95% CI = [0.04, 0.21]; Figure 2A) and less decline in GCA over time (b = 0.12, 95% CI = [0.03, 0.22]; Figure 2B). In these models, baseline cognition was not associated with change in the number of sleep disorders over time. Bivariate relationships between sleep disorders and cognition (both MoCA and GCA) were not found for participants with MCI. However, for participants with dementia, having a greater number of sleep disorders at baseline was related to better MoCA performance at baseline (b = 0.11, 95% CI = [0.02, 0.19]; Supplementary Figure 1). Results from the reduced models were consistent with those from the full models (Supplementary Table 5).

Figure 2.

Bivariate associations between number of sleep disorders and cognition in participants with impaired cognition. Participants with impaired cognition include participants with both mild cognitive impairment and dementia. Significant paths at p < .05 are indicated by solid lines; bolded solid lines represent significant bivariate associations; nonsignificant paths are indicated by dotted lines. Path coefficients represent standardized parameter estimates with 95% confidence intervals in parentheses. Path coefficients are from the fully adjusted models including all sociodemographic and health-related covariates based on bootstrapped models with 5,000 bootstrapped samples. GCA = general cognitive ability; MoCA = Montreal Cognitive Assessment. (A) Bivariate associations between number of sleep disorders and scores on the Montreal Cognitive Assessment. (B) Bivariate associations between number of sleep disorders and general cognitive ability.

Supplemental Analyses

Discrepancy between % of symptoms versus % of diagnoses

On average, 16% and 14% of participants had symptoms of sleep apnea and insomnia at baseline, respectively. Among those who reported symptoms of sleep apnea, 90.33% of the unimpaired cognition group, 91.96% of the MCI group, and 87.85% of the dementia group had clinician-identified sleep apnea. Among those who reported the symptoms of insomnia, 84.18% of the unimpaired cognition group, 85.11% of the MCI group, and 72.92% of the dementia group had clinician-identified insomnia.

Cognition by self-reported symptoms of sleep disorders

Results testing the associations of symptoms of sleep apnea and insomnia with cognition are presented in Supplementary Table 6. For participants with unimpaired cognition, baseline performance or change in cognition for either MoCA or GCA did not differ by baseline symptoms of sleep apnea and insomnia. For participants with impaired cognition, baseline MoCA performance was better (b = 0.83, 95% CI = [0.25, 1.41]) for those reporting symptoms of sleep apnea, and both baseline MoCA (b = 1.42, 95% CI = [0.76, 2.08]) and baseline GCA (b = 0.13, 95% CI = [0.01, 0.25]) were better for those reporting symptoms of insomnia. The decline in MoCA scores over time was less steep (b = 0.59, 95% CI = [0.23, 0.95]) for those with versus without insomnia.

Participants with MCI who reported symptoms of insomnia at baseline had less decline in MoCA over time (b = 0.53, 95% CI = [0.18, 0.89]) and better GCA performance at baseline (b = 0.16, 95% CI = [0.02, 0.30]). Participants with dementia who had symptoms of sleep apnea (b = 1.66, 95% CI = [0.72, 2.59]) and insomnia (b = 1.64, 95% CI = [0.49, 2.79]) at baseline had better MoCA performance at baseline. Results from the reduced models were consistent with those from the full models (Supplementary Table 7).

Discussion

This study presents novel insights into how sleep disorders relate to cognitive aging. Based on the literature that reports the individual associations of sleep apnea or insomnia with cognitive decline (de Almondes et al., 2016; Ferman et al., 1999; Incalzi et al., 2004; McCarter et al., 2016; Shi et al., 2018) and co-occurrence of sleep disorders in older adults (Guarnieri et al., 2012; Suzuki et al., 2017; Yaffe et al., 2014), we expected that having each of sleep apnea or insomnia or a greater number of sleep disorders across multiple types would be associated with greater cognitive decline over time. Our findings, however, show that older adults with cognitive impairment have a better cognitive trajectory if they have sleep apnea, insomnia, or a greater number of multiple sleep disorders at baseline. Although these findings are surprising and contrary to our expectations and the majority of the literature, our results are in line with a few studies (Hegde et al., 2020; Zawar et al., 2022) that also found better cognitive function and brain health in those with clinical diagnoses of these sleep disorders. One possible explanation for the complex pattern of findings may be SDoH and social-ecological model of sleep and health (Grandner, 2019; Majoka & Schimming, 2021), which suggest that societal/neighborhood influences (e.g., access to healthcare) may have downstream consequences on how individuals with a sleep disorder diagnosis have better cognition over time. Challenges to the diagnosis of sleep disorders in older adults, especially in those with cognitive impairment, selection bias, and lack of information on the history of sleep treatment may also be potential confounding factors. There are several important points from this study that provide implications for future research and clinical practice.

First, our results show that the prevalence of symptoms of sleep apnea or insomnia is quite low in the older population. In our sample, symptoms of sleep apnea or insomnia reported by participants or coparticipants (a family member or a close friend) were 16% and 14%, respectively. About 90% of those self-reporting sleep apnea symptoms and 82% of those self-reporting insomnia symptoms were clinically diagnosed, with a lower rate of clinical diagnosis in the dementia group. It is well known that demented persons experience a variety of sleep and other dysfunctions due to the dementia itself. Perhaps, clinicians may pay less attention to sleep problems in patients with dementia because those are implicated in dementia pathology. However, part of it may relate to challenges in the clinical diagnosis of sleep disorders, particularly in those with cognitive impairment.

In this study, cognitive evaluations were based on standardized clinical evaluation, but evaluations of sleep disorders were based on a questionnaire asking participants whether they have ever had a sleep disorder in four prevalent types (i.e., sleep apnea, hyposomnia or insomnia, RBD, or other). As sleep disorder diagnoses were not verified through additional assessments, there might be variations by clinicians and challenges to the diagnosis of sleep disorders especially in individuals with cognitive impairment (Brzecka et al., 2018). Only some of clinical assessment tools for screening and diagnosing sleep disorders have been validated in older subjects and nearly none in geriatric patients or in individuals with dementia (Frohnhofen et al., 2020). Thus, there may be those with hidden or underdiagnosed sleep disorders in a more severely impaired cognition group, which might have contributed to the unexpected relationship between the presence or a greater number of sleep disorder diagnoses and less cognitive decline. Alternatively, it is possible that those with cognitive impairment underreport sleep conditions, leading to discrepant findings.

Finally, those with clinical diagnoses of cognitive impairment and sleep disorders may represent a selected group (e.g., higher income) who had access to better healthcare and were potentially referred to clinics for proper diagnosis and treatment, which might have contributed to better baseline cognition and less cognitive decline observed in this study. Even with clinical diagnosis, however, the proportion of older adults who have managed sleep disorder (through a safe treatment with a provider) may still be low. Studies show that about one third of patients with moderate to severe OSA do not receive continuous positive airway pressure (CPAP) treatment (Nogueira et al., 2018), which is the gold standard of sleep apnea treatment. The proportion of people with insomnia who receive an evidence-based treatment like cognitive behavioral therapy for insomnia (CBT-I) may be even lower, due to the limited number of trained providers and patient financial burden (Arnedt et al., 2021). Given that treatment of sleep disturbance (subjective symptoms measured by questionnaire) may reduce the risk of future probable AD (Burke et al., 2018), future efforts should be focused more on identifying hidden/underdiagnosed sleep disorders in older adults with cognitive impairment to delay their further cognitive decline. Training more clinicians in sleep medicine, offering telemedicine sleep diagnosis services for those living in disadvantaged or rural communities, and including sleep assessments (through comprehensive scales and/or objective measures) in regular healthcare checkups may be helpful. It may also be important to make sleep treatments like CPAP and CBT-I more affordable to individuals with lower socioeconomic status.

The association between sleep disorders and cognition differed by participants’ cognitive conditions, which may provide insights for future intervention programs. For those with unimpaired cognition, sleep disorders were not related to cognition, probably because of the lower prevalence of sleep disorders in this group compared to those with impaired cognition. For this group, intervention programs that help maintain their cognitive function (e.g., Nicholson et al., 2022) may be helpful to prevent the onset or increase of sleep disorders and associated risk for cognitive impairment. For those with MCI, there was no significant bivariate association between sleep disorders and cognition, although the beta coefficients were in the same direction with those found in the dementia group. This is intriguing and requires more research attention, because participants with MCI were more likely to have sleep disorders than participants with dementia as well as than those with unimpaired cognition (Table 1). Our supplemental results showed that participants with MCI who reported symptoms of insomnia at baseline had better cognition and less decline in cognition over time. Similarly, participants with dementia who had symptoms of sleep apnea and insomnia at baseline had better cognition at baseline. Perhaps, reporting symptoms of sleep problems may also relate to the individual’s cognitive ability of sensing changes in sleep. Challenges and barriers to the diagnosis of MCI have been reported (Judge et al., 2019; Purser et al., 2006). Our results add that individuals with MCI may also have difficulty describing their sleep problems and potentially underreport their symptoms, which may hinder detecting a significant risk factor. Finally, for those with dementia, having sleep apnea, insomnia, or a greater number of multiple sleep disorders at baseline was related to better cognition at baseline. Sleep disorders in individuals with dementia are often manifested as nocturnal waking and wandering behavior, disrupting the sleep quality of family caregivers (Oliveira et al., 2019; Waligora et al., 2019). Thus, there is a critical need to improve the detection and diagnosis of sleep disorders in individuals with dementia.

Limitations and Future Directions

This study has limitations that may guide future research. First, the presence of sleep disorders was identified by clinicians based on participant (or coparticipant) self-report of whether the subject had clinician-diagnosed sleep disorders. Future work may need to incorporate objective assessments of sleep to verify sleep disorder diagnoses. Also, the treatment history of the diagnosed sleep disorders was not collected and symptoms of sleep disorders were collected only at baseline. Being able to account for treatment would improve generalizability of our findings. In addition, keeping treatment constant across participants could have arguably strengthened our links between true, untreated sleep disorders and cognitive outcomes. Therefore, it may be that we offer conservative estimates of the associations between sleep and cognition. Future research may consider measuring the symptoms, clinical diagnosis, and treatment history of each sleep disorder at each measurement timepoint to better understand how the trajectory of cognitive aging differs by underdiagnosed/untreated sleep disorders. It is also possible that cognitive aging may be accelerated by increase in the number of sleep disorders and/or by extended time with untreated sleep disorders. Future work could examine potential nonlinear change in cognition (e.g., quadratic, cubic) by sleep disorders over time. Second, the sample was highly educated and mostly White, thus findings may not generalize to those with less education and racial minorities. Further, the NACC data set may not be representative of the U.S. population (Arce Rentería et al., 2023), which warrants caution in interpreting findings for older adults in the United States. Moreover, cognitive measures used in this study (i.e., MoCA and GCA) may be less than ideal. Especially, MoCA may be less sensitive to measure cognitive function in those with unimpaired cognition. Future research may want to incorporate more diverse measures of cognition to precisely capture changes in cognition across the spectrum of cognitive status. Lastly, a specific mechanism underlying the association between sleep disorders and cognitive decline was not tested in this study. The SDoH and social-ecological model of sleep and health perspectives (Grandner, 2019; Majoka & Schimming, 2021) suggest several potential mechanisms that may explain our finding (e.g., lack of healthcare access). Future research may need to examine the putative underlying mechanisms specifically.

Conclusion

This study contributes to better understanding the complex relationship between sleep disorders and cognitive aging. Sleep disorders are prevalent in geriatric population, and only a small proportion of them are identified by clinicians. This is problematic because there may be hidden or underdiagnosed sleep disorders that may differentially be related to cognitive status. By capturing multiple sleep disorder diagnoses, examining their longitudinal associations with cognition, and making specific inferences by cognitive groups, this study suggests that more detection and treatment of sleep disorders may delay further cognitive decline in those with cognitive impairment. Future work may need to focus more on early detection of potential sleep disorders and offering proper treatments to those with a higher risk of pathological aging.

Funding

The current study was supported by award numbers R56AG065251 and R01HL163226 (PI S. Lee, PhD) and RF1AG056331 (PI M. Wallace, PhD) from the National Institutes of Health [NIH]/NIA. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIA or the NIH. R. Andel was funded by project number LX22NPO5107 (MEYS): financed by European Union—Next Generation EU. M. Nelson was funded by NIH/NIA (award number F31AG077865). Outside of the current work, O. M. Buxton discloses that he received subcontract grants to Penn State from Proactive Life (formerly Mobile Sleep Technologies) and doing business as SleepSpace (National Science Foundation [grant #1622766] and NIH/National Institute on Aging Small Business Innovation Research Program [R43AG056250, R44 AG056250]). The NACC database is funded by NIA/NIH grant U24 AG072122. NACC data are contributed by the NIA-funded ADRCs: P30 AG062429 (PI James Brewer, MD, PhD), P30 AG066468 (PI Oscar Lopez, MD), P30 AG062421 (PI Bradley Hyman, MD, PhD), P30 AG066509 (PI Thomas Grabowski, MD), P30 AG066514 (PI Mary Sano, PhD), P30 AG066530 (PI Helena Chui, MD), P30 AG066507 (PI Marilyn Albert, PhD), P30 AG066444 (PI John Morris, MD), P30 AG066518 (PI Jeffrey Kaye, MD), P30 AG066512 (PI Thomas Wisniewski, MD), P30 AG066462 (PI Scott Small, MD), P30 AG072979 (PI David Wolk, MD), P30 AG072972 (PI Charles DeCarli, MD), P30 AG072976 (PI Andrew Saykin, PsyD), P30 AG072975 (PI David Bennett, MD), P30 AG072978 (PI Neil Kowall, MD), P30 AG072977 (PI Robert Vassar, PhD), P30 AG066519 (PI Frank LaFerla, PhD), P30 AG062677 (PI Ronald Petersen, MD, PhD), P30 AG079280 (PI Eric Reiman, MD), P30 AG062422 (PI Gil Rabinovici, MD), P30 AG066511 (PI Allan Levey, MD, PhD), P30 AG072946 (PI Linda Van Eldik, PhD), P30 AG062715 (PI Sanjay Asthana, MD, FRCP), P30 AG072973 (PI Russell Swerdlow, MD), P30 AG066506 (PI Todd Golde, MD, PhD), P30 AG066508 (PI Stephen Strittmatter, MD, PhD), P30 AG066515 (PI Victor Henderson, MD, MS), P30 AG072947 (PI Suzanne Craft, PhD), P30 AG072931 (PI Henry Paulson, MD, PhD), P30 AG066546 (PI Sudha Seshadri, MD), P20 AG068024 (PI Erik Roberson, MD, PhD), P20 AG068053 (PI Justin Miller, PhD), P20 AG068077 (PI Gary Rosenberg, MD), P20 AG068082 (PI Angela Jefferson, PhD), P30 AG072958 (PI Heather Whitson, MD), andP30 AG072959 (PI James Leverenz, MD).

Conflict of Interest

O. M. Buxton received honoraria/travel support for lectures from Boston University, Boston College, Tufts School of Dental Medicine, Eric H. Angle Society of Orthodontists, Harvard Chan School of Public Health, and Allstate; received consulting fees from SleepNumber; and received an honorarium for his role as the Editor in Chief of Sleep Health (sleephealthjournal.org). M. L. Wallace discloses that she receives funds as a statistical consultant for Sleep Number Bed, Health Rhythms, and Noctem Health, outside the current work.

Data Availability

The NACC data are available to other researchers through data request at https://naccdata.org/. Analytic methods specific to the current study are available upon request from the corresponding author. The current study was not preregistered with an analysis plan in an independent, institutional registry.

Acknowledgments

The authors thank Dr. Walter Kukull for his review and feedback on the manuscript.

Author Contributions

Soomi Lee (Conceptualization [Lead], Funding acquisition [Lead], Investigation [Lead], Methodology [Supporting], Supervision [Lead], Validation [Lead], Writing—original draft [Lead], Writing—review & editing [Lead]), Monica Nelson (Formal analysis [Lead], Methodology [Lead], Visualization [Lead], Writing—original draft [Equal], Writing—review & editing [Equal]), Fumiko Hamada (Investigation [Equal], Writing—original draft [Equal], Writing—review & editing [Equal]), Meredith Wallace (Funding acquisition [Supporting], Investigation [Equal], Methodology [Supporting], Supervision [Equal], Writing—review & editing [Equal]), Ross Andel (Formal analysis [Supporting], Funding acquisition [Supporting], Investigation [Equal], Methodology [Equal], Supervision [Equal], Validation [Supporting], Writing—review & editing [Equal]), Orfeu Buxton (Funding acquisition [Supporting], Investigation [Equal], Supervision [Equal], Validation [Supporting], Writing—review & editing [Equal]), David Almeida (Funding acquisition [Supporting], Investigation [Equal], Supervision [Equal], Writing—review & editing [Equal]), Constantine Lyketsos (Data curation [Lead], Investigation [Equal], Methodology [Supporting], Resources [Supporting], Supervision [Equal], Validation [Equal], Writing—review & editing [Equal]), and Brent Small (Formal analysis [Supporting], Methodology [Supporting], Supervision [Equal], Validation [Equal], Writing—review & editing [Equal])