Sleep Disturbances Predict Cognitive Decline in Cognitively Healthy Adults

Abstract

Background:

The effect of nighttime behaviors on cognition has not been studied independently from other neuropsychiatric symptoms.

Objective:

We evaluate the following hypotheses that sleep disturbances bring increased risk of earlier cognitive impairment, and more importantly that the effect of sleep disturbances is independent from other neuropsychiatric symptoms that may herald dementia.

Methods:

We used the National Alzheimer’s Coordinating Center database to evaluate the relationship between Neuropsychiatric Inventory Questionnaire (NPI-Q) determined nighttime behaviors which served as surrogate for sleep disturbances and cognitive impairment. Montreal Cognitive Assessment scores defined two groups: conversion from 1) normal to mild cognitive impairment (MCI) and 2) MCI to dementia. The effect of nighttime behaviors at initial visit and covariates of age, sex, education, race, and other neuropsychiatric symptoms (NPI-Q), on conversion risk were analyzed using Cox regression.

Results:

Nighttime behaviors predicted earlier conversion time from normal cognition to MCI (hazard ratio (HR): 1.09; 95% CI: [1.00, 1.48], p = 0.048) but were not associated with MCI to dementia conversion (HR: 1.01; [0.92, 1.10], p = 0.856). In both groups, older age, female sex, lower education, and neuropsychiatric burden increased conversion risk.

Conclusion:

Our findings suggest that sleep disturbances predict earlier cognitive decline independently from other neuropsychiatric symptoms that may herald dementia.

INTRODUCTION

Sleep disturbances, including insomnia, poor sleep quality, and sleep fragmentation, are implicated in the development and progression of Alzheimer’s disease (AD) and related dementias in older adults [1–7]. Proposed mechanisms include an association between sleep disturbances and reduced clearance of amyloid-β from the brain [8–11] as well as circadian rhythm disturbances to accelerated neurodegeneration [12].

Sleep disturbances in those at risk for dementia often do not appear in isolation. Neuropsychiatric symptoms have a strong association with cognitive impairment [13–17] and are known predictors of progression of cognitive decline among patients with pre-existing dementia [18]. A systematic review found mixed results in the interactions between sleep regulation and cognition [19], suggesting that comorbid neuropsychiatric symptoms often preclude clear results [19]. These findings highlight the need to identify reliable parameters to measure the relationship between sleep disturbances and cognition. Additionally, they identify a knowledge gap about how this relationship is influenced by the presence of neuropsychiatric symptoms such as depression and anxiety.

The Neuropsychiatric Inventory Questionnaire (NPI-Q) is a validated screening tool for assessing neuropsychiatric symptoms among cognitively impaired individuals [20, 21] and includes a sleep disturbance item [18, 22, 23]. The sleep disturbances item in the NPI-Q, referred to as “nighttime behaviors,” is defined as “interruptions of sleep and daytime sleepiness.” We used the prospective National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS), a large longitudinal database, to evaluate the following hypotheses: 1) nighttime behaviors bring increased risk of earlier cognitive impairment, and 2) the effect of nighttime behaviors is independent from other neuropsychiatric symptoms that may herald dementia. These risks are important to clarify since interventions for sleep problems may attenuate or slow cognitive decline in susceptible individuals.

MATERIALS AND METHODS

Database

The NACC UDS contains prospective de-identified data collected under the National Institute on Aging’s Alzheimer’s Disease Center (ADC) program to facilitate collaborative research in Alzheimer’s disease and related dementias [24–27]. Thirty-nine ADC sites have their own participant eligibility criteria, and all are required to contribute to the NACC UDS. Informed consent was obtained at the respective centers. Data from visits conducted between September 2005 and December 2021 were included in the analysis. All participants with a diagnosis of dementia at initial visit were excluded from analysis.

Primary measures

Nighttime behaviors measured the presence or absence of sleep disturbance. This clinician-evaluated variable contains the response to the question, “Does the patient awaken you during the night, rise too early in the morning, or take excessive naps during the day?” This variable is based on the co-participants interview (i.e., spouse, children, or other caregivers). This measure of sleep disturbances was validated as significantly associated with the severity of cognitive impairment and biomarkers associated with Alzheimer’s Disease in the cross sectional analysis [28].

The Montreal Cognitive Assessment (MoCA) score was used to assess the severity of the cognitive deficit [29]. During the course of the NACC data collection, the MoCA replaced the Mini-Mental Status Examination. To provide a homogeneous score of cognition, the database variable NACCMMSE was converted using a validated conversion nomograph [30]. To create cognitive categories, MoCA cut-off scores of dementia < 20, 20≤MCI≤25, and normal > 25 were used [31].

To measure the effects of nighttime behaviors on the conversion to and progression of cognitive impairment, we defined two groups based on the MoCA cut-off scores described above.

Conversion from normal to MCI included those who scored within the normal MoCA range at the initial visit and converted to MCI (versus those who remained normal) during follow-up.

Conversion from MCI to dementia included those who scored as MCI at the initial visit and converted to dementia (versus maintenance of MCI) during follow-up.

Predictors of conversion

To account for the cumulative prognostic effect of other neuropsychiatric symptoms, we created a covariate, the NPI weight (“NPIweight”). We defined NPIweight as the sum of the positive answers on the NPI-Q survey, excluding the “nighttime behaviors” variable, divided by the number of valid answers in the NPI-Q. The symptoms of the NPI-Q (excluding nighttime behaviors) included agitation, anxiety, apathy, appetite problems, delusions, depression/dysphoria, disinhibition, elation, hallucinations, irritability, and/or motor disturbances. The inclusion of NPIweight allowed us to evaluate the effect of nighttime behaviors (and its underlying sleep disturbances) separate from other neuropsychiatric symptoms while preserving degrees of freedom in the statistical model. Like nighttime behaviors, the rest of NPI-Q variables were also based on co-participants or the caregiver’s interview.

Age (years, determined at the initial visit), sex (male versus female), race (dichotomized to white and non-white), and education (total years) were utilized as other covariates because older age, female sex, lower education level and non-white race are well-established risk factors for dementia [32–34].

Statistical analysis

Data summarization

Baseline categorical scaled clinical characteristics were summarized by frequencies (n) and relative frequencies (%), while continuous scaled and integer scaled baseline characteristic were summarized by the median, interquartile range (IQR), and range of the empirical distribution.

Univariate analyses

Univariate relationships between nighttime behaviors and the remaining baseline characteristics (i.e., age at initial visit, sex, race, education, and NPI-weight), were examined by using the Fisher-exact test for the nighttime behavior versus sex, and the nighttime behaviors versus race comparisons and using the two sample Wilcoxon Rank Sum test for the nighttime behaviors versus age at initial visit, the nighttime behaviors versus education, and the nighttime behaviors versus NPIweight comparisons.

Cox multivariate regression analyses

Using the patients’ baseline information for nighttime behaviors, age at initial visit, education, NPIweight, sex and race, two independent Cox multivariate regression analyses were conducted. One regression analysis was focused on examining if there are unique associations between the time to conversion from normal cognitive function to MCI and nighttime behaviors, age at initial visit, education, NPIweight, sex, and race, while the other regression analysis focused on examining if there are unique associations between the time to conversion from MCI to dementia and nighttime behaviors, age at initial visit, education, NPIweight, sex, and race.

The predictor variable “nighttime behaviors” was specified as a categorical variable with levels: “yes” and “no”. “Age at initial visit” was specified as a categorical variable with levels:<40 years, [40–50 years], [50–60 years], [60–70 years], [70–80 years], [80–90 years], and > 90 years. “Education” was specified as an integer scaled continuous variable. “NPIweight” was specified as continuous scaled variable. “Sex” and “race” were specified as categorical variables with levels “male”/“female” and “white”/“non-white”, respectively.

With regard to time-to-event censoring, in the Cox regression analyses for time to conversion, normal cognition to MCI conversion times and MCI to dementia conversion times for those patients who failed to convert were treated as right censored observations.

In terms of hypothesis testing, the type-III Wald Chi-square statistic served as the hypothesis testing pivotal quantity. Per Wald Chi-square test, the null hypothesis was stated in the conventional Cox multivariate regression null hypothesis testing framework, i.e., where under the null hypothesis it is assumed that there is no unique association between the instantaneous risk for the event (e.g., time to conversion from normal cognitive function to MCI) and the level/value of the predictor variable (e.g., nighttime behaviors). Rejection of the null hypothesis implies that there is a unique association between the instantaneous risk of the event and the level/value of the predictor variable. The null hypothesis rejection criterion was set a priori at the 0.05 significance level for all of the type-III Wald Chi-square tests.

The proportional hazards assumption was tested via the Therneau and Grambsh proportional hazards tests [35].

Missing data

Due to the censoring aspect of the outcome data, we did not attempt to impute missing time to event data. For the “conversion time from normal cognitive function to MCI” Cox multivariate regression analysis, 4,232 (26.7%) patients were missing conversion time information, and of the 11,656 patients who had conversion time information (5,621 right censored times, 6,035 non-censored times), 11,074 (95.0%) patients (5,357 right censored times, 5,717 non-censored times) had complete predictor variable information. For the “conversion time from MCI to dementia” Cox multivariate regression analysis, 4,215 (33.6%) patients were missing conversion time information, and of the 8,332 patients who had conversion time information (4,794 right censored times, 3,538 non-censored times), 7,927 (95.1%) patients (4,548 right censored times, 3,379 non-censored times) had complete predictor variable information. Due to the small percentage of patients who had missing predictor variable information among those with conversion time information, we did not impute missing predictor variable information.

RESULTS

Baseline characteristics

Baseline clinical characteristics of the two groups are summarized in Table 1A and 1B. Table 1A shows the associations between nighttime behaviors and baseline characteristics of the normal to MCI group, while Table 1B shows the associations between nighttime behaviors and baseline characteristics of the MCI to dementia group. Of the 15,568 patients in the normal to MCI group, 2,229 (14.3%) endorsed nighttime behaviors at the initial visit (Table 1A). The median age of patients with nighttime behaviors at initial visit was 69.0 years (IQR: [63.0, 75.5 years]). 56.4% of the patients were female (n = 1,258), and 87.6% were white (n = 1,953).

Table 1A.

Univariate comparisons of baseline characteristics: age, education, NPIweight (an adjusted sum of the neuropsychiatric inventory questionnaire (NPI-Q) score), race, and sex by presence or absence of nighttime behaviors on the initial visit of patients assessed in the NACC database for (A) the sample of those who tested in the normal range of cognition according to Montreal Cognitive assessment at initial visit and (B) those who tested with minimal cognitive impairment (MCI) at initial visit. Empirical distributions for age at first visit, education, and NPIweight are summarized by the median and interquartile range (IQR), and range of the empirical distribution and between nighttime behavior comparisons were conducted by way of the Wilcoxon Rank Sum test. Empirical frequency distributions for race and sex are summarized as frequencies (n) and relative frequencies (%) and between nighttime behavior comparisons of race and sex were conducted by way of the Fisher’s exact test. A. Normal to MCI

	Nighttime behaviors	N	Median [IQR]	Range [Min, Max]	p
Age at first visit (y)	Present	2,229	69.0 [63.0, 75.5]]	[21.0, 97.0]	<0.001
	Absent	13,339	70.0 [64.0, 76.0]	[18.0, 104.0]
Education (y)	Present	2,222	16.0 [14.0, 18.0]	[10.0, 28.0]	0.001
	Absent	13,298	16.0 [14.0, 18.0]	[0, 29.0]
NPI weight	Present	2,179	0.18 [0.00, 0.27]	[0.00, 1.00]	<0.001
	Absent	12,887	0.00 [0.00. 0.09]	[0.00, 0.91]
		n	Nonwhite (n, %)	White (n, %)	p
Race	Present	2,229	276 (12.4%)	1,953 (87.6%)	0.001
	Absent	13,339	2002 (15.0%)	11,337 (85.0%)
			Male (n, %)	Female (n, %)
Sex	Present	2,229	971 (43.6%)	1,258 (56.4%)	<0.001
	Absent	13,339	4728 (35.4%)	8,611 (64.6%)

Table 1B.

B. MCI to dementia

	Nighttime behaviors	N	Median [Interquartile Range]	Range [Min, Max]	p
Age at first visit (y)	Present	2,636	72.0 [65.0, 78.0]	[21.0, 99.0]	<0.001
	Absent	9,645	73.0 [67.0, 79.0]	[18.0, 102.0]
Education (y)	Present	2,627	16.0 [12.0, 18.0]	[0, 26.0]	0.483
	Absent	9,618	16.0 [13.0, 18.0]	[0, 30.0]
NPI weight	Present	2,578	0.27 [0.09, 0.46]	[0, 1.00]	<0.001
	Absent	9,350	0.00 [0.00, 0.18]	[0, 0.91]
		n	Nonwhite (n, %)	White (n, %)	p
Race	Present	2,636	459 (17.4%)	2,177 (82.6%)	<0.001
	Absent	9,645	2,349 (24.4%)	7,296 (75.6%)
			Male (n, %)	Female (n, %)
Sex	Present	2,636	1,448 (54.9%)	1,188 (45.1%)	<0.001
	Absent	9,645	4,330 (44.9%)	5,315 (55.1%)

Of the 12,281 patients in the MCI to dementia group, 2,636 (21.5%) endorsed nighttime behaviors at the initial visit (Table 1B). The median age of patients with nighttime behaviors at initial visit was 72.0 years (IQR: [65.0, 78.0 years]). 45.1% (n = 1,188) were female, and 83% (n = 2,177) were white. In general, patients who presented with nighttime behaviors on the initial visit were significantly younger, disproportionately male and disproportionately white. Those with nighttime behaviors also significantly presented with a significantly greater neuropsychological burden (higher NPIweight). Those with nighttime behaviors, in the normal to MCI group, at the initial visit had an average of two other positive NPI-Q symptoms; those without nighttime behaviors usually had no other symptoms in the NPI-Q survey. Comorbid neuropsychiatric symptoms were even more frequent in the MCI to dementia group. Those with nighttime behaviors had about three other NPI-Q symptoms and those without nighttime behaviors had about one other neuropsychiatric symptom. These findings indicate that those with nighttime behaviors also presented with a larger burden of other neuropsychiatric symptoms, or in other words, with more advanced clinical features of disease.

Nighttime behaviors and conversion from normal cognition to MCI

After adjusting for other predictors, the instantaneous risk for conversion from normal cognitive function to MCI was uniquely associated with nighttime behaviors (Table 2A, Fig. 1). The instantaneous risk for conversion from normal cognitive function to MCI was 1.09 time greater (95% CI: [1.00, 1.18], p = 0.048) for those patients who at their initial visit reported experiencing nighttime behaviors than for those patients who at their initial visit did not report experiencing nighttime behaviors. Those with nighttime behaviors who were cognitively normal at the initial visit converted to MCI about 12.1 months earlier than those without nighttime behaviors at the sample median (Corresponding median times to MCI conversion are provided in Supplementary Table 1).

Table 2A.

Cox-multivariate regression model type-III Wald Chi-square ANOVA summary and hazard ratio estimates for comparing the instantaneous risk of conversion to MCI for the sample of patients who tested in the normal range of cognition according to the Montreal Cognitive Assessment at the initial visit (A, B), and for comparing the instantaneous risk of conversion from MCI to dementia for those who tested with minimal cognitive impairment (MCI) according to the Montreal Cognitive Assessment at the initial visit (C, D). Table 2A lists the Cox multivariate regression model type-III Wald Chi-square ANOVA results for testing the null hypothesis that there is no unique association between the predictor variable and time to conversion to MCI. Table 2B lists the Cox multivariate regression model hazard ratios for comparing the instantaneous risk of conversion to MCI. Table 2C lists the Cox multivariate regression model type-III Wald Chi-square ANOVA results for testing the null hypothesis that there is no unique association between the predictor variable and time to conversion from MCI to dementia. Table 2D lists the Cox multivariate regression model hazard ratios for comparing the instantaneous risk of conversion from MCI to dementia. A-B: Conversion from Normal to MCI

Predictor Variable	Degrees of Freedom	Wald Type III Chi-Square Statistic	p
Nighttime Behavior	1	3.91	0.048
Age at Visit #1	6	791.09	<0.001
Educations	1	153.20	<0.001
NPIweight	1	204.35	<0.001
Race	1	241.65	<0.001
Sex	1	206.67	<0.001
Model	11	1601.60	<0.001

Fig. 1.

Kaplan Meire curves for the cumulative probability for no conversion to MCI by Nighttime Behavior (A), age at visit 1(B), years of education (C), NPIweight (D), race (E) and sex (F). Corresponding median times to conversion to MCI are provided in Supplementary Table 1.

After adjustment for all remaining predictor variables, increasing age, lower education, greater NPIweight, non-white race, and female sex categories were uniquely associated with greater risks of MCI conversion (Table 2B).

Table 2B.

Predictor Variable	Ratio	Hazard Ratio [95% CI]	Type III Wald Chi-Square Statistic	p
Nighttime Behavior	Yes: No	1.09 [1.00, 1.18]	3.91	0.048
Age at Visit #1	40–50:<40	1.90 [1.11, 3.26]	5.42	0.020
	50–60:<40	2.31 [1.40, 3.80]	10.86	0.001
	60–70:<40	3.36 [2.06, 5.50]	23.47	<0.001
	70–80:<40	5.16 [3.16, 8.43]	43.07	<0.001
	80–90:<40	7.73 [4.72, 12.65]	66.20	<0.001
	>90:<40	12.06 [6.98, 20.86]	79.47	<0.001
Education†	X years: X + 1 years	1.06 [1.05, 1.07]	153.20	<0.001
NPI Weight‡	X + 0.1: X	1.16 [1.14, 1.18]	204.35	<0.001
Race	Non-white: White	1.75 [1.63, 1.88]	241.65	<0.001
Sex	Female: Male	1.49 [1.41, 1.58]	206.68	<0.001

^†X denotes the X years of education and X + 1 denotes X + 1 years of education.

^‡denotes a NPIweight of X, and X + 0.1 denotes a NPIweight of X + 0.1.

Diagnostic analysis revealed that the Cox model proportional hazards assumption held for all Cox regression terms other than sex (p = 0.034).

Nighttime behaviors and conversion from MCI to dementia

After adjusting for other predictors, the instantaneous risk for conversion from MCI to dementia was not associated with nighttime behaviors (Table 2 C). The instantaneous risk for conversion from MCI to dementia was 1.01 times greater (95% CI: [0.92, 1.10], p = 0.856) for those patients who at their first visit reported experiencing nighttime behaviors than for those patients who at their first visit did not report experiencing nighttime behaviors.

Table 2C.

C-D: Conversion from MCI to Dementia

Predictor Variable	Degrees of Freedom	Wald Type III Chi-Square Statistic	p
Nighttime Behavior	1	0.03	0.855
Age at Visit #1	6	152.22	<0.001
Educations	1	43.93	<0.001
NPIweight	1	364.25	<0.001
Race	1	0.41	0.521
Sex	1	4.35	0.037
Model	11	608.98	<0.001

After adjustment for all remaining predictor variables, increasing age, lower education, greater NPIweight, non-white race, and female sex categories were uniquely associated with greater risks of MCI conversion (Table 2D).

Table 2D.

Predictor Variable	Ratio	Hazard Ratio [95% CI]	Type III Wald Chi-Square Statistic	p
Nighttime Behavior	Yes: No	1.01 [0.92, 1.10]	0.03	0.856
Age at Visit #1	40–50:<40	1.61 [0.67, 3.89]	1.13	0.289
	50–60:<40	2.10 [0.94, 4.73]	3.23	0.072
	60–70:<40	2.11 [0.95, 4.72]	3.33	0.068
	70–80:<40	2.83 [1.27, 6.31]	6.47	0.011
	80–90:<40	3.53 [1.58, 7.89]	9.47	0.002
	>90:<40	4.60 [2.01, 10.53]	12.99	<0.001
Education†	X years: X + 1 years	1.04 [1.03, 1.05]	43.93	<0.001
NPI Weight‡	X + 0.1: X	1.21 [1.19, 1.23]	364.25	<0.001
Race	Non-white: White	0.97 [0.89, 1.06]	0.41	0.521
Sex	Female: Male	1.08 [1.00, 1.16]	4.35	0.037

^†X denotes the X years of education and X + 1 denotes X + 1 years of education.

^‡denotes a NPIweight of X, and X + 0.1 denotes a NPIweight of X + 0.1.

Diagnostic analysis revealed that the Cox model proportional hazards assumption held for all Cox regression terms.

DISCUSSION

Our findings demonstrate that the presence of sleep disturbances, defined in the NACC database as nighttime behaviors encompassing frequent awakening during the night, early awakening, and daytime sleepiness, predicted higher risk of conversion from normal cognition to MCI. More importantly, because our statistical model accounted for other neuropsychiatric symptoms, our results indicate that nighttime behaviors predicted earlier cognitive decline independent of other neuropsychiatric symptoms.

The present study supports our earlier findings that the cross-sectional prevalence of sleep disturbances increases with worsening cognition [28]. Insomnia, sleep fragmentation, shorter sleep time, poor sleep quality and excessive daytime sleepiness all have been associated with poor cognitive outcomes and increased dementia risk [1–5, 36–38]. The present study extends these associations by statistically modeling the longitudinal risk of cognitive conversion over a consistently maintained large sample of patients seen in long-term follow-up.

Our selection of variables is an important consideration in the evaluation of dementia risk factors. We used the nighttime behaviors variable in the NACC database as our primary marker for sleep disturbances. Our previous work established that nighttime behaviors identified on the initial visit were associated with worse cognition, smaller hippocampal volumes, and higher levels of CSF biomarkers [28] and that “nighttime behaviors” are a more informant marker of sleep disturbances compared to some of the other more subjective tools of insomnia assessment in this patient population [28]. The “nighttime behaviors” variable assesses sleep-wake cycle, daytime sleepiness, and wakefulness at night. Nighttime behaviors, as assessed by the NPI-Q, may be useful in the clinical setting as it is an efficient and easy yet reliable way of assessing sleep disturbances.

The NACC database contains complementary variables for the clinical determination of MCI or dementia. We used MoCA scores as our cognitive measurement because it provided consistent, uniform and objective, cognitive measurements important in the construction of survival curve models. Various ADCs participate in NACC. Variability in assessment by different clinicians and the center’s practice may lead to a more subjective assessment of cognitive status. Despite its limitations, a score-based assessment of cognitive status such as MoCA or MMSE is likely to be longitudinally stable and reliable overtime because it utilizes the same objective questionnaire and is less dependent on clinician judgment. Moreover, MoCA scores in NACC database are adjusted for patient’s education. The use of MoCA scores also allowed us to synchronize our findings with our earlier study from the same database that determined that MoCA scores, in terms of sleep variables available in the database, returned meaningful relationships with dementia biomarkers [28].

Potential mechanisms by which sleep disturbances may play a role in cognitive impairment include circadian rhythm disturbances, and alteration in homeostatic and motivational processes [12, 39–41]. Sleep dysregulation may result from a variety of factors; some aspects of the variable “nighttime behaviors” can be interpreted as representing disruption or dysregulation of sleep-wake cycles. Irregular sleep-wake rhythm disorder, prevalent in dementia, is a loss of a circadian pattern in sleep-wake cycles and presents as nighttime sleep fragmentation and daytime sleep [42]. Both human and animal studies have shown that circadian rhythm disturbances occur in dementia, maybe associated with increased risk of incident dementia and may even appear years before onset of cognitive decline [40]. Animal models have identified a possible role of circadian dysfunction in dementia and worsening neurodegeneration [12]. Thus, it is possible that circadian rhythms beyond sleep-wake states may play a role in pathology in dementia patients [40]. Interactions between sleep regulation and neurodegeneration are likely bi-directional. Although patients with normal cognition who presented with nighttime behaviors converted earlier to MCI—which may imply cause-and-effect—earlier subclinical neurodegenerative processes could have been present. We propose that the nighttime behaviors were not predictive in those who presented with MCI because sleep and cognition were already mutually intertwined, and the clinical sign of nighttime behaviors was not sensitive enough to predict earlier conversion to dementia.

Neuropsychiatric symptoms are prevalent in patients with cognitive impairment and as many as 75% of dementia patients may have one or more neuropsychiatric symptoms [15, 43]. In the present study, we controlled for the effect of combined neuropsychiatric symptoms by the use of the variable NPIweight, since the number of positive responses on the NPI-Q was a strong predictor of cognitive worsening. Our study supports earlier work that showed that patients with an initial NPI-Q score greater than zero (endorsement of any symptom[s]), experienced a 1.37 times greater risk of conversion to dementia [44] than those with no symptoms of the NPI-Q. While higher NPI-Q scores have been shown to predict cognitive decline [23] and mortality among the cognitively impaired [45], nighttime behaviors variable has only been studied in a handful of studies [46, 47]. And the role of nighttime behaviors, independent of other neuropsychiatric symptoms, in predicting cognitive decline has not been studied previously. We extend earlier work by demonstrating that nighttime behaviors at the initial visit have a significant effect on conversion risk beyond those of a symptom complex assessed by the NPI-Q score. The importance of the effect of sleep disturbances lies in that sleep disturbances represent symptoms that may be easily modified, and thereby possibly attenuate the severity or delay the onset of dementia.

We found older age, female sex, lower education, and non-white race to be risk factors for conversion to dementia and MCI. These findings are consistent with previous literature [17, 32–34, 48]. Similar to some of the previous studies [3, 5], in our cohort, men were more likely to endorse nighttime behaviors at the initial visit [3, 5]. Thus, even though men were more likely to present with nighttime behaviors, females were more likely to convert to worsening cognition.

Limitations of our study lie in the measurement and operationalization of sleep disturbances as available in the NACC UDS. The assessments of sleep available in the NACC database are not physiological measurements of sleep disturbances, and they bear the limitations of any survey of clinical history. Our preliminary study determined, however, that the “nighttime behaviors” variable was the most informant proxy of sleep disturbances in the NACC UDS because it fit well with the expected relationship between sleep and cognition within this patient population and was associated with biomarkers of dementia such as hippocampal volume loss and CSF biomarkers [28]. However, further studies are needed to validate nighttime behaviors against direct measures of sleep and to evaluate what physiological changes in sleep accompany the label of “nighttime behaviors” as gathered in the NACC database. NPI-Q is a brief questionnaire that does not provide a comprehensive assessment or severity of insomnia or sleep disturbances. Future studies may find that a specific tool like the Insomnia Severity Index [49] may provide a better assessment of sleep disturbances. The full score of the NPI-Q battery (range 0–12) was not always available, and the missing data for individual questions may have affected our results. Although a large database such as the NACC UDS provides a rich data source, it could mean that statistically significant findings may carry unclear clinical significance. Moreover, our analyses did not include the range of all possible comorbid medical conditions that could affect cognitive decline, such as diabetes mellitus or hypertension, because the incomplete documentation of these combined conditions severely decreased the number of subjects with complete data available for analyses. Medications which could affect sleep or were prescribed to treat sleep disturbances, such as psychotropic or sedative hypnotics, were not documented reliably in the NACC database. Another notable limitation of our study is that the response to nighttime behaviors variable and the other NPI-Q questions was provided by the caregiver. Therefore, these reflect subjective assessments based on caregiver perception. However, as cognitive impairment progresses, the patient with dementia may not be able to accurately report neuropsychiatric symptoms. Finally, another limitation of our work is that NACC UDS participants do not comprise a nationally representative sample, which may affect generalizability of our results.

In summary, we provide additional evidence for an association between sleep disturbances and risk of cognitive impairment. While neuropsychiatric symptoms have a strong association with cognitive impairment, our findings identify nighttime behaviors as an independent predictor of cognitive decline.

Early identification of sleep disturbances via routine screening in cognitively healthy and cognitively at-risk adults may have significant implications. Measures to treat sleep disturbances may counter the deleterious effects of sleep disturbances on maintenance of cognition and delay the onset of cognitive decline. Early intervention and aggressive management of sleep disturbances, therefore, may present a path to mitigation of cognitive decline.

Fig. 2.

Kaplan Meier curves for the cumulative probability for no conversion from MCI to dementia by Nighttime Behavior (A), age at visit 1 (B), years of education (C), NPIweight (D), race (E), and sex (F). Corresponding median times to conversion to MCI are provided in Supplementary Table 2.

DATA AVAILABILITY

De-identified data for this article is available via National Alzheimer’s Coordinating Center.