Association of Sleep Disturbances and Cognitive Decline among Cognitively Normal Elders Over Time

Background: While sleep disturbances (SD) has been shown to be associated with worse cognition, but the causal relationship between the two subjects to debate. Our objective was to investigate the longitudinal impact of SD on cognitive function. Objective: To determine the effect of self-reported clinical diagnosis of SD on longitudinal changes in brain amyloid-PET, CSF-biomarkers (Aβ42, T-tau and P-tau) and cognitive function in cognitively normal. Methods: A total of 463 cognitively normal elders (357 normal and 106 SD) were included. Alzheimer’s Disease Neuroimaging Initiative (ADNI) participants were collected from 2005 to 2020. The generalized linear mixed models adjusting variables which were selected by the Akaike Information Criterion (AIC) and the marginal effect estimation method was used to estimate the risk effect of SD. Cox proportional hazards regression models estimated the relative hazard of Alzheimer’s Disease (AD), among baseline SD patients. Results: The age range of participants was 73.60±5.71 years old, and the female proportion was 43.63%. In adjusted regression models, Participants with baseline SD had higher likelihood of developing worse cognition over subsequent follow-up, PACC (decrease 7.53 points [95%CI, 7.36-7.70]; P<0.001), MMSE (decrease 5.26 points [95%CI, 5.17-5.35]; P<0.001), and CDR–Sum of Boxes (increase 5.61 points [95%CI, 5.67-5.54]; P=0.001). Similarly, Cox regression analysis suggested that sleep disturbances is a risk factor of AD (HR=1.55, 95% CI=1.08 to 2.22). Conclusion: SD probably is a warning sign of AD, because it is associated with greater likelihood of cognitive decline or dementia over time. Associations are likely multifactorial and could be explained by intervening variables in the path from SD to dementia, or by common risk factors for pathological processes in brain. These ndings suggest need for more attentions of older adults with sleep compromise.


Introduction
Sleep disturbances (SD) has been de ned as di culty in initiating or maintaining sleep, along with an impairment of daytime functioning [1]. The elderly have been associated with increased night sleep disturbances whose prevalence is estimated to be approximately 50% [2,3]. Sleep disturbances in this population has adverse effects including an increased risk of dementia, hypertension, diabetes, obesity, depression, heart attack, and stroke [4,5]. Alzheimer's disease (AD) is a dementia syndrome in which cognitive impairment interferes with the performance of daily activities. The daily activities that are affected include sleep (sleep disturbances) which is the most common in patients with AD [6]. Consequently, whether sleep disturbances, which is common in the elderly, is the cause of AD or not should be veri ed with more evidences.
Accumulating evidence suggests that sleep disturbances precede the neurobiology of AD and symptoms of cognitive decline by years. Poor sleep quality was associated with Mild Cognitive Impairment (MCI), which often represents a prodromal phase of preclinical dementia [7]. Moreover, it is thought that sleep disturbances starts to occur in the preclinical phases of AD, as they have been found to predict incident dementia [8][9][10]. However, these studies don't use the people with normal cognition and overlook the variance of cognition from time, and so it is hard to say in this case that the main reason of cognitive decline is the effect of time or the sleep disturbance. Moreover, there is still no consensus on whether sleep disturbances could cause cognitive decline. Some studies report there is no association between sleep disturbances and cognition [11]. Furthermore, some studies even report slightly better cognitive functioning in those with sleep disturbance [12]. So, there is a con ict about whether sleep disturbances are the pre-symptomatic stage of AD or not.
Preclinical AD is important for the understanding of ageing and AD and drug development because of the hypothesis that disease-modifying interventions will be most effective when initiated early [13]. If we can ensure that sleep disturbances in the preclinical AD have adverse effects on subsequent cognitive change, then we can know the causal relationship between sleep and dementia and whether it's essential to slow the progression of AD by regulating sleep disorder. The purpose of this current study was to compare the differences in subsequent cognitive change between normal sleep and sleep disturbances at baseline among normal cognition or subjective memory concerns individuals.

Study Participants
Data used were collected from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). ADNI was launched in 2005 as a public-private partnership, led by Michael W. Weiner, MD [14]. The primary purpose of ADNI was to investigate whether serial MRI, PET, other biomarkers, also clinical and neuropsychological assessment could be combined to assess the progression of MCI and early dementia. In this current study, we downloaded and merged the data (16 years follow-up, from 2005 to 2020) which was updated at 30th Jun 2020 from the o cial website of ADNI, including 83 ADNI sites in the Canada and United States.
For the present study, a subset of individuals with Normal Cognition (NC) or subjective memory concerns those who were self-reported or reported by a study partner or clinician and was included in the analyses. Individuals with sleep disturbances were termed as SD based on the medical history diagnosed by clinicians with "insomnia", "sleep disorder", "sleep apnea", and "sleep disturbance". As shown in Figure S1, the participants without the medical history information of sleep disturbances were considered to have normal sleep and were termed as Normal Sleep (NS). So, we got 1165 individuals: 152 SD, 1013 NS. After that, we excluded the participants without baseline information of CSF Aβ42, CSF tau, CSF phospho-tau (p-tau), medical history, variables of demographic, medication history, and cognition evaluated variables.
Lastly, we got 463 individuals: 106 SD, 357 NS. The follow-up time in months for the individuals of this study is shown in Table S1.

Standard Protocol Approvals, Registrations, and Patient Consents
The data in this study was approved by institutional review boards of all ADNI centers. Additionally, patient informed consent has also been done by ADNI.  [19]. Logical Memory Delayed Recall scores are based on years of education and were required to be at least 9 for 16 years of education, at least 5 for 8 to 15 years of education, and at least 3 for 0 to 7 years of education (0 to 25, worst to best). Additionally, accumulating studies suggest that increased amyloid protein precedes the cognitive symptoms of AD [20]. So, we required that the participants should have the CSF β-amyloid peptide (Aβ 42 ) and the pro les related with Aβ 42 including CSF tau, p-tau. The above variables needed participants to be followed up at least one time or more.

Confounding variables
Age is associated with the sleep disturbance, so it was treated as the covariate in our model. Additionally, the other variables like APOEε4 carriage, gender, family history of dementia, family history of AD, medication of AD, education level, and ventricular volume at baseline were considered into the model. The intaking of medicine against AD was collected from concurrent medications log. Race and ethnicity were self-reported as required by the National Institutes of Health. ADNI was approved by the institutional review boards of all participating institutions. Written informed consent was obtained from all participants at each site.

Statistical Analysis
The pro les of sleep disturbances and normal sleep group were described and the differences between these groups were tested by Pearson χ 2 and two-sample t-test. Progression by sleep disturbances group among individuals with normal cognition was analyzed by generalized linear mixed effects (GLME) models which could control the bias from individual and confounding variables. The interaction of time and sleep disturbances was also included in the model. The Cox regression was also used to investigate if the sleep disturbances could lead to AD to happen in the follow-up years. The primary GLME model was used to estimate the longitudinal effect of baseline SD over time. Firstly, this analysis treated the time as the continuous variable, which could show the long-term effect of SD to cognitive normal individuals in this cohort study. Although time is measured discretely, there are enough numerical values that we could t a line to it, which could consider the spacing between each time point. Additionally, the time was treated as a categorical variable in the other models of which the design is same as the primary model, which could show the short-term effect of SD in each follow-up. The Akaike Information Criterion (AIC) [21], a model selection tool, was used to decide if quadratic terms for time should be added. The other variables like age, APOEε4 carriage, gender, family history of dementia, family history of AD, medication of AD, education level, and ventricular volume at baseline were selected into the model by AIC. Moreover, the structure of the model was also chosen by AIC from random intercept, random slope, compound symmetric, and unstructured option.
For the estimation of interaction between the sleep disturbances and time we used the marginal effect estimation method. The marginal effect estimation [22] could evaluate the effect of sleep disturbances on cognitive pro les at each follow-up year. So, we could see how the interaction of sleep disturbances and time in uence the cognitive decline. Considering fewer data were available at the later study visits, we emphasized the point estimation at 6 year.

Sensitivity Analyses
Considering possible reverse causality, sensitivity analyses were used to explore robust associations between sleep disturbances and cognitive decline. People with elevated CSF Aβ42 are more likely to get cognitive decline than those with low level CSF Aβ42. So, we reanalyzed the association between sleep disturbances and cognition function by excluding participants whose CSF Aβ42 less than 985pg/mL [23]. All of the above analyses and graphs were performed with R software (version 4.0.2, https://www.Rproject.org/). All the statistical tests were two-sided and P value < 0.05 were considered as statistically signi cant.

Results
As can be seen from Table 1, all essential characteristics were used to compare between NS (n = 357) and SD (n = 106). The proportion of male SD was signi cantly lower than that of NS. Whereas, the percentage of family histories of patients with Dementia, family histories of patients with AD, and subjective memory concern in SD was higher than that in NS with statistical signi cance. The participants in SD had fewer education years than those with normal sleep. Additionally, comparing the pro les of cognition between SD and NS, we found that the people with sleep disturbances have a higher value of PACC, logical memory delayed recall, FDG, and the lower of ADAS13. That indicated people with sleep disturbances have better cognition in the baseline. Table S2, Table S3, and Table S3 shows that sleep disturbance, time, and the interaction of sleep disturbances and time are signi cantly associated with PACC, MMSE, and CDR-Sum of Boxes when the model adjusted confounding factors selected by AIC. Although there were protective effects of sleep disturbances for PACC, MMSE, and CDR-Sum of Boxes at baseline, the sleep disturbances could be a risk factor for these cognitive outcomes by the years. As shown in Fig. 1, when the time was treated as a continuous variable, there was a decreased tendency by years on PACC and MMSE, and the CDR-Sum of Boxes score rose dramatically by time. After rst year follow-up, the difference in scores for cognitive variables between sleep disturbances and normal sleep was larger and larger. There was not this kind of relationship in the categorical time models. So, sleep disturbances made the PACC, MMSE, and CDR-Sum of Boxes get worse by years in the continuous time model.
According to Table S9, sleep disturbance, time, and the interaction of sleep disturbances and time are signi cantly associated with CSF Aβ42 in continuous time models. Although time and sleep disturbances alone were not considered risk factors for Aβ42 protein elevation, the interaction of sleep disturbances and time did contribute to Aβ42 protein elevation. The effect of time made all of cognitive pro les get worse signi cantly. As shown in Fig. 3, the concentration of Aβ42 protein dropped sharply by years and the concentration in sleep disorder group was higher than that in the normal group after 1 year in the continuous time model.    Mean neurodegeneration pro les are depicted in Table S7 and Fig. 3. The percentage of ICV for ventricular volume increased gradually by years and that of sleep disturbances group was higher than the normal one in the continuous time model. As shown in Table S8, S10, and S11, there was not this kind of relationship between sleep disturbances and CSF tau, p-tau, and hippocampal volume in models. The effect of time made neurodegeneration pro les get worse signi cantly but for hippocampal volume. Figure 4 shows the hazard ratio of sleep disturbances for the AD in the Cox model adjusting other covariates. The hazard ratio APOEz4 allele was 2.18 (1.54, 3.07), which is the highest one among these variables, followed by sleep disturbances and male. 1.55 (1.08, 2.22) was the second number, belonged to sleep disturbance, and the male was the lowest one with 1.53 (1.05, 2.23). The above results were statistically signi cant (P < 0.05).
In the sensitivity analysis that excluded participants whose CSF Aβ42 more than 985pg/mL, the results did not change substantially. We found that sleep disturbances can still have a negative effect on the PACC, MMSE, and CDR-Sum of Boxes, CSF Aβ42, and ventricular ICV volume over years. Moreover, we reanalyzed the COX regression for AD, which depict sleep disturbances can still be the risk factor of AD.

Discussion
In this longitudinal study of cognitively normal participants, participants with sleep disturbances were easier to develop cognitive decreased symptoms compared participants with normal sleep. The cognitive decreased symptoms included the downward trend of PACC and MMSE, and the upward trend of CDR-Sum of Boxes score, CSF Aβ42, and ventricular volume in 10 to 14 years follow-up. Additionally, people with sleep disturbances who got AD were 1.55 (1.08, 2.22) times more than that of those with normal sleep. That suggests that sleep disturbances probably represent the pre-symptomatic stage of AD.
There were some differences in demographic and cognitive pro les between people with sleep disturbances and normal sleep in the baseline. The sleep disturbances accounted for a higher male percentage than that of those having normal sleep. However, prior studies [24,25] found women to be more frequent and explain about sleep disturbances saying that women are easier to be in uenced by stress and depression which make them di cult to get a good sleep. In this study, the average age of people was around 73 and men got more diseases in this kind of age than women[26], which makes them get di cult to fall in sleep. That's why there were more males with sleep disturbances than females with sleep disturbance. Meanwhile, the proportion of family histories of patients with dementia or AD in sleep disturbances was higher than that in the group with normal sleep, but the percentage of individuals with APOEz4 allele was no signi cant difference between these two groups. More interesting, the cognitive pro les including PACC, logical memory delayed recall, FDG, and ADAS13 at sleep disturbances group were better than that at normal sleep. So, whether sleep disturbances has adverse effects on cognitive decline or other factors or not needs the longitudinal research part in this study to clarify.
In the longitudinal research part, time was treated as a continuous variable and categorical variable respectively with the generalized linear mixed model to investigate how sleep disturbances in uences the subsequent cognitive change. The signi cant variables in baseline were selected by AIC as covariates into the model to ensure that these variables could in uence the cognition. We found that the subsequent cognition outcomes in sleep disturbances group get signi cantly worse than that in the normal group,  [29] which evaluates the performance of cognitive questionnaires suggested that the CDR-Sum of Boxes is more precise in measuring the severity of cognitive dysfunction than the MMSE and we also found that sleep disturbances in uences that outcome. Consequently, sleep disturbances in baseline probably could make the cognition subsequently decline.
For outcomes of markers of amyloid, sleep disturbances showed the long-term increased effect on the Aβ in CSF. There was not any in uence from sleep disturbances on some neurodegeneration markers (CSF tau and p-tau). That is supported by the amyloid hypothesis [30] of Alzheimer's disease which supposes that the concentration of Aβ would increase rst and then the high concentration of Aβ induce the amyloid plaques and neuro brillary tangles. Later, the concentration of tau and p-tau would be at a high level. Our ndings are consistent with this hypothesis, when the CSF Aβ rise in individuals with sleep disturbance, the CSF tau and p-tau haven't any signi cant change. Consequently, sleep disturbances could be a preclinical stage to AD.
The Cox regression analysis showed sleep disturbances is a signi cant risk factor of AD (HR = 1.55, 95% CI = 1.08 to 2.22) considering other covariates, which is supported by previous studies. Similarly, a multicentre longitudinal study [31] demonstrated that late-life sleep disturbances is associated with an increased dementia risk (HR = 1.94, 95% CI = 1.08 to 3.49). Osorio et al. [32] found that individuals with sleep disturbances are easier to get AD. Sterniczuk et al. [10] reported that participants who have trouble in sleeping will get dementia or AD within 4 years. The follow-up time of this study cohort was longer than that of previous studies, and the follow-up time was 10 to 14 years. Besides, the confounding factors were fully considered in this research paper, therefore the results were more reliable and credible.
There are some strengths in this study. First of all, the individuals in this longitudinal research study all had a normal cognition, which could rule out the effects of cognitive decline on sleep disturbance. Secondly, we used the marginal effect estimation which quanti es the interaction between sleep disturbances and time on cognitive decline. Additionally, the generalized linear mixed model was used in this research, which could consider the bias of individual-level and use the data which doesn't meet the requirements of the traditional model. Finally, a lot of covariates were included in this study and were selected by AIC, which could exclude the probability that cognitive decline is caused by other factors.
There are also some limitations to this research. Firstly, the number of participants in our study is limited, therefore the relationship between cognitive decline and sleep disturbances needs a large sample size cohort to verify. Next, the lack of people with clinical imaging indicators in the late follow-up resulted in a short follow-up time and small sample size at each follow-up. Lastly, the research study didn't consider to use biological experiments to verify the results. However, we plan to address these questions in further study.

Conclusion
In conclusion, this study con rmed that sleep disturbances in the baseline is associated with subsequent cognitive decline among cognitively normal elders and is an increased risk of AD. Sleep disturbances probably is a pre-symptomatic stage of AD, which will be important for import by the clinical doctor to prevent people to get cognitive decline or AD. Individuals with sleep disturbances should pay more attention to their cognition and be given active treatments. Further studies should explore whether there are some pieces of evidence from the biological mechanism to support the relationship between sleep disturbances and cognitive decline.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and material
All data are available in the ADNI database (ida.loni.usc.edu). Derived data is available from the corresponding author on request by any quali ed investigator.

Competing interests
The authors have no competing interests to declare.

Funding
This current study was nancially supported by The Program of National Natural Science Foundation of China (Serial Number: 81903408) (XN Wang), The National Key R&D Program of China (Serial Number: 2016YFC1302804) (XH Guo). The funding was neither was used for the study design nor data collection but to cover for the publication fees.

Authors' contributions
Feng Wei and Xiuhua Guo designed this research study, did the statistics analyses, and drafted this paper. Mandela William Nzoyoum Kuetche, Meng Zhang, and Mengmeng Liu collected the data from ADNI, revised the manuscripts, and designed the gures in this paper. Deginet Aklilu gave suggestions to the generalized linear mixed model, revised the method part, and interpretation of data for the work in this research. Wei Wang , Xiaonan Wang, and Xiuhua Guo revised the work of whole manuscript, and offered contributions to the conception of the work from epidemiology and statistics view.