Social Network and Risk of Poor Sleep Outcomes in Older Adults: Results from a Spanish Prospective Cohort Study

Introduction

Sleep disorders are a common health problem with prevalence estimates ranging from 10% to 50% of the general population.¹ Individuals with sleep problems frequently report physical and mental fatigue, altered mood, and poor concentration, which have a negative impact on their family life, professional activities and social relationships.² Several meta-analyses of observational studies have shown that poor sleep duration (<6 or >9 hours/day) is associated with an increased risk of stroke,³ coronary heart disease,^4,5 hypertension,⁵ diabetes,^3,5 obesity,^3,5 falls,⁶ all-cause^3–5 and cancer-specific^7,8 mortality.

The prevalence of social isolation among older adults is also high. According to recent estimates, only in the United States (US), 7.7 million community-dwelling older adults report being socially isolated.⁹ A social network is defined as the structure of all relevant relationships that link individuals with the people surrounding them. Prospective studies have shown an increased mortality among individuals with weaker social relationships.¹⁰ A lack of social support has also been associated with a higher risk of anxiety and depression,¹¹ cardiovascular disease (CVD)¹² and diabetes,¹³ and a lower adherence to medications for chronic disease management and control.^14–16

Although there is evidence that not having social support is associated with poor sleep outcomes,¹⁷ most prospective studies are based on social support in the workplace, do not account for baseline sleep characteristics or do not assess sleep duration.¹⁷ Moreover, the few prospective studies on social support and sleep characteristics among the elderly have been carried out in Anglo-Saxon and Asian populations,^18–23 with no previous study having evaluated this association in an older Mediterranean population. This is important because there are cultural differences in how people seek, receive and perceive social support from their social network,^24,25 and some of these differences may moderate the relationship between social interactions and health outcomes.²⁶ For example, older adults in Mediterranean countries are less frequently involved in volunteer groups or part-time employments, and rely more on the contact with family members than those from other developed regions.^27,28 Some studies have also shown that older adults from more individualistic societies²⁹ and societies with better welfare regimes like those in Northern and Central Europe, are less likely to suffer from loneliness and its consequences than those in more collectivistic cultures with lower levels of social protection in Southern and Eastern Europe.³⁰ Also, when compared to older adults from other European regions, the Spanish elderly have longer nighttime sleep, shorter daytime naps, and later bed and wake up times.³¹ Given these particularities, we aimed to evaluate whether social network was prospectively associated with changes in sleep quality and sleep duration over time in community-dwelling older adults in Spain.

Methods

Study Participants and Design

The Seniors-ENRICA cohort was established during 2008–2010 with individuals selected by multi-stage stratified random sampling from the non-institutionalized Spanish population aged ≥60 years (wave 1).³² For these analyses, we used data from 2519 subjects participating in wave 2 (2012), the first wave with social network information available. At baseline (2012) and follow-up (2015), information regarding sociodemographic variables, sleep characteristics, health behaviors and morbidity was collected. During follow-up (2012–2015), 207 participants died and 488 were lost to follow-up (please see Supplementary Figure 1). Socio-demographic, lifestyle, and clinical characteristics of individuals lost to follow-up and those remaining were similar.³³ This study was conducted in accordance with the Declaration of Helsinki. Study participants gave written informed consent, and the study was approved by the Clinical Research Ethics Committee of La Paz University Hospital, Madrid.

Study Variables

At baseline (2012), social network was assessed with eight questions about marital status, cohabitation, frequency of contacts with relatives other than those with whom you live, frequency of contacts with friends or neighbors, amount of time spent alone, availability of someone to go for a walk with, receipt of emotional support (ie, having someone to talk, share feelings or problems with), and receipt of instrumental support (ie, having someone who can take care of important things in case of sickness).³⁴ Using this information, the following dichotomous (0/1) variables were created: not being married, living alone, not having daily/almost daily contacts with family members other than those living with, not having daily/almost daily contact with friends or neighbors, spending more than 8 hours/day alone, lacking someone to go for a walk with, not having emotional support, and not having someone to take care of important things in case of sickness. An overall social network index (SNI) was calculated as the sum of points for each dichotomous item and ranged from 0 (better) to 8 (worse).³⁵

The SNI showed an acceptable internal consistency (Cronbach’s alpha, 0.68). It has content validity because the questions covered different aspects of the structure (ie, marital status, frequency of contact with relatives), and function (ie, emotional help, instrumental help) of social relationships within the network.³⁶ With regards to construct validity, there is empirical evidence that the questions used enable subject´s principal social ties to be assessed.³⁷ Finally, our measure of social network has criterion validity, because most of its items have been shown to be associated with both the physical and social dimensions of health-related quality of life, as measured by the Short Form-36 (SF-36) questionnaire on a sample of 3600 persons representative of the Spanish population aged ≥60 years.³⁸

In 2012 and 2015, information on sleep duration and on symptoms of sleep disturbance was collected. Sleep duration was obtained with the question: “Approximately, for how long do you usually sleep per night?” Responses were categorized as short (<6 h), normal (7–8 h), or long sleep (>9 h).³⁹ A sleep disturbance scale was built as the sum of the points given to each symptom of sleep disturbance (ie, bad overall sleep, difficulty falling asleep, awakening during nighttime, early awakening with difficulty getting back to sleep, use of sleeping pills, feeling restless in the morning, being asleep at daytime, and having an Epworth Sleepiness Scale (ESS) >10), with higher scores meaning worse sleep (please see Supplementary Table 1 for further details).⁴⁰ Individuals were then classified according to quartiles of the sleep disturbance scale, with those in the highest quartile (ie, those with ≥4 points) being deemed to suffer from poor sleep quality. The main advantage of quantiles is that they help characterize the shape of the relationship between exposure and outcome without a priori assumptions about the form of the relationship and ensuring equal size groups (which increases the power of the analyses).

In 2012, information was collected on age, sex, education (≤primary, secondary and university studies), smoking status (current, former and never smoker), alcohol consumption (never/occasional drinkers, moderate drinkers, heavy/binge drinkers and ex-drinkers), and time (hours/week) in different leisure-time activities (watching television [TV], at the computer, reading, and listening music).⁴¹ Physical activity (PA) was assessed with the validated questionnaire used in the EPIC-cohort study in Spain, and expressed in metabolic equivalents of tasks (METs)- hours/week.⁴² Recreational PA included walking (commuting, shopping, or leisure time), cycling (commuting or leisure time), and playing sports (running, playing soccer, doing aerobics, swimming, or playing tennis), while household PA included doing household chores (cleaning, cooking, doing laundry or children rearing), gardening, and doing home repair (do-it-yourself activities). Compared with the rate of energy expended at rest, the assigned METs were 2.5 for walking and household chores and 4.0 for cycling, playing sports, gardening, and home repair.⁴³ Weight and height were measured with standardized procedures, and the body mass index (BMI) was calculated as weight in kilos divided by squared height in meters. Food consumption was collected with a validated diet history and caffeine intake (mg/d) estimated from standard composition tables of foods in Spain.⁴⁴ Adherence to a Mediterranean-type diet was assessed with the Mediterranean Diet Adherence Screener (MEDAS) score, which ranges from 0 (lowest) to 14 (highest adherence to the Mediterranean diet) and consists of 2 questions on food intake habits (“sofrito” and olive oil) considered characteristic of the Spanish Mediterranean diet, and 12 questions on food consumption frequency.⁴⁵

Depression was ascertained with the 10-item version of the Geriatric Depression Scale (GDS-10),⁴⁶ with the following yes-no questions being considered: 1) “Are you basically satisfied with your life?” 2) “Have you dropped many of your activities and interests?” 3) “Do you feel that your life is empty?” 4) “Are you afraid that something bad is going to happen to you?” 5) “Do you feel happy most of the time?” 6) “Do you often feel helpless?” 7) “Do you feel you have more problems with your memory than most?” 8) “Do you feel full of energy?” 9) “Do you feel that your situation is hopeless?” and 10) “Do you think that most people are better off than you are?”. The total number of “yes” responses to questions 2, 3, 4, 6, 7, 9, and 10, and “no” responses to questions 1, 5, and 8 were added to estimate a total depressive symptom score ranging from 1 to 10. Psychological distress was estimated using the 12-item general health questionnaire (GHQ-12), a screening tool that assesses the respondent’s current psychological state and asks if that differs from his/her usual state.⁴⁷ The GHQ-12 included the following questions: “Have you recently … ” 1)“Lost much sleep over worry?” 2)“Felt constantly under strain?” 3)“Felt you couldn´t overcome your difficulties?” 4)“Been feeling unhappy or depressed?” 5)“Been losing confidence in yourself?” 6)“Been thinking of yourself as a worthless person” 7)“Been able to concentrate on what you´re doing” 8)‘Felt that you are playing a useful part in things’ 9)“Felt capable of making decisions about things?” 10)‘Been able to enjoy your normal day-to-day activities’ 11)“Been able to face up your problems” and 12)“Been feeling reasonable happy, all things considered?”. The response options were worded in terms of “less than usual”, “no more than usual”, “rather more than usual”, or “much more than usual”. Questions 1 through 6 were scored 1 point if participants responded “rather more than usual”, or “much more than usual”, whereas questions 7 through 12 were scored 1 point if they answered “less than usual” or “no more than usual”. The final scale ranged from 0 to 12, with higher scores indicating higher short-term psychological distress. Health-related quality of life was measured using the second version of the 12-item Short Form Health Survey, previously validated in Spain.⁴⁸ Participants were also asked about any prescribed drugs, and these were checked by the study staff against drug packages at home. Finally, in 2012, participants reported if they had experienced ≥ 2 falls in the last year, and if they had suffered from any of the following physician-diagnosed diseases: CVD (ischemic heart disease, stroke, heart failure), diabetes, hypertension, cancer, respiratory disease (chronic bronchitis, asthma), or osteomuscular disease (osteoarthritis, arthritis).

Statistical Analyses

From the 1824 individuals who participated in 2015, we excluded 153 without data on sleep variables at baseline or at follow-up, and 227 without data on social network or potential confounders at baseline, leading to a final analytical sample of 1444 individuals (Supplementary Figure 1).

Differences in the social network by sociodemographic, lifestyle, morbidity, and sleep variables were tested using analysis of variance (ANOVA) for continuous variables and Pearson’s chi square for categorical variables. In cross-sectional analyses we evaluated the association of the SNI (as continuous and categorized into quartiles) and its components with the prevalence of short (<6) and long (>9) nighttime sleep, as well as with the prevalence of having ≤1, 2, 3 or ≥4 sleep disturbance indicators. Results from these models are presented as relative prevalence ratios (RPR) and their 95% confidence intervals (95% CI). In prospective analyses, we evaluated the association of the SNI (as continuous and categorized into quartiles) and its components with: a) changes in nighttime sleep duration (continuous) between 2012 and 2015; b) risk of short (<6) or long (>9) nighttime sleep in 2015; c) changes in the number of indicators of sleep disturbance (continuous) between 2012 and 2015; and d) risk of poor sleep quality (≥4 indicators of sleep disturbance). Model 1 adjusted for age, sex, education, and baseline values of the studied outcomes; model 2 further adjusted for baseline information on lifestyle factors (tobacco smoking, alcohol drinking, time/day spent in different activities [ie, watching TV, reading, listening music, at the computer], recreational and household PA, diet quality, caffeine consumption, and BMI); as well as for baseline GHQ-12 and GDS-10 scores, baseline number of drug treatments, prevalence of chronic morbidities, and number of falls during the preceding year. Results are presented as mean changes [MC] (for continuous outcomes) or odds ratios [OR] (for categorical outcomes), along with their 95% CI.

Finally, we evaluated whether the studied associations varied with sex, age, education, previous diagnosis of depression, PA, diet quality and caffeine consumption, by using likelihood-ratio tests that compared models with interaction terms and without. Given that no significant interactions were found, results are presented for the total sample.

Statistical significance was set at 2-sided p<0.05. Analyses were performed with STATA (version 13.0; Stata Corp.)

Results

Among study participants, 29% were not married, 20% lived alone, 29% did not have frequent contact with family members other than those with whom they lived, 40% did not have frequent contact with friends or neighbors, 19% spent more than 8 hours/day alone, 41% lacked someone to go for a walk with, 6% lacked emotional support and 49% instrumental support. Supplementary Table 2 shows the frequency of poor social network items stratified by the participants’ characteristics. Women, older participants, those with lower education, never drinkers or smokers, more sedentary and less active participants, as well as those with greater morbidities more frequently lived alone and spent >8 hours during daytime alone. Younger men showed lower contact with family members and a greater lack of instrumental support than their counterparts.

In cross-sectional analyses, a 1-point increase in the SNI was associated with having long night-time sleep (RPR and 95% CI: 1.17 [1.03–1.33]) and presenting ≥4 sleep disturbance indicators (RPR: 1.12; 95% CI: 1.00–1.25), while showed a borderline association with short night-time sleep (RPR: 1.08; 95% CI: 0.99–1.19) (Supplementary Table 3). Unmarried participants showed the highest prevalence of short sleep (RPR: 1.41; 95% CI: 1.00–1.19).

Over a mean follow-up of 2.8 years, participants lost an average of 10.45 minutes of sleep (SD: 75.9) (please see Supplementary Table 4), and a total of 88 and 104 participants developed short and long sleep, respectively (Table 1). Regarding the mean number of sleep disturbance indicators, this was unchanged, as some individuals improved while others worsen their sleep. However, of 1200 individuals with good sleep quality at baseline, 127 (10.6%) developed ≥4 sleep disturbance indicators (Table 2).

Table 1 Prospective Association Between Values of the Social Network Index (SNI) and Its Components in 2012, Changes in Nighttime Sleep Duration Between 2012 and 2015, and Risk of Short (<6 Hours) or Long (>9 Hours) Nighttime Sleep in 2015

Table 2 Prospective Association Between Values of the Social Network Index (SNI) and Its Components in 2012, Changes in the Sleep Disturbance Scale Between 2012 and 2015, and Risk of Developing ≥4 Poor Sleep Disturbance Indicators in 2015

After multivariate adjustment, and compared to individuals in the lowest (best) quartile of baseline SNI, those in the second, third and fourth quartiles, respectively, suffered MC (95% CI) in sleep duration of 2.32 (−7.58–12.22), −2.70 (−13.19–7.79) and −13.04 (−23.41 - −2.67) minutes from 2012 to 2015; p for trend=0.02. Additionally, a higher risk of sleeping <6 h/d at follow-up was observed in those with worse social network (OR per 1 item increase in the SNI: 1.22; 95% CI: 1.05–1.42; Table 1). Living alone, spending ≥8 h/d alone, not having someone to go for a walk with, and not having emotional support showed the strongest associations with reductions in sleep duration and risk of short sleep.

Table 2 shows no overall association between baseline SNI scores and changes in the number of sleep disturbance indicators (MC in sleep disturbance scale per 1 item increase in the SNI: 0.03; 95% CI: −0.00–0.07). However, an increased risk of developing ≥ 4 sleep disturbance indicators at follow-up was observed among those with worse baseline social network (OR per 1 item increase in the SNI: 1.13; 95% CI: 1.00–1.30). When each sleep disturbance indicator was considered individually, a worse SNI score was only significantly associated with the risk of “early awakening and difficulty getting back to sleep” (odds ratio per 1-item increase: 1.20; 95% CI: 1.07–1.35), and no mean changes in the EES scores between baseline and follow-up were observed (data not shown in tables).

Discussion

In this cohort of community-dwelling older adults from a Mediterranean country, a worse SNI was cross-sectionally associated with the prevalence of long-term sleep and prospectively associated with a reduction in sleep duration and an increased risk of short sleep over time, regardless of other relevant sociodemographic, behavioral, and clinical covariates. Living alone, spending ≥8 h/d alone, not having someone to go for a walk with, and not having emotional support showed the strongest associations with reductions in sleep duration and risk of short sleep. Participants with worse SNI, and in particular those lacking someone to go for a walk with and those lacking emotional support, were more likely to display ≥ 4 poor sleep quality indicators at baseline, as well as to develop ≥ 4 sleep disturbance indicators when these were not present at baseline.

Despite sleep disorders being more common among older compared with younger adults,⁴⁹ and the evidence showing that older adults with both short and long sleep display worse cognitive function,^50,51 and are at increased risk of depression,⁵² falls,⁵³ pain symptoms⁵⁴ and disability,⁵⁵ only a few previous studies have evaluated the prospective association between measures of social network and sleep characteristics in community-dwelling older adults.^18–23 Findings from these cannot be directly compared with ours since they considered different social relationships and used different measures of sleep quality and duration, but they also showed a positive influence of social networks on sleep outcomes. Specifically, a study on 447 older adults followed for 2 years found a prospective relationship between emotional loneliness, as measured with 3 items of the de Jong Gierveld Scale, and poor sleep quality, assessed with the Pittsburgh Sleep Quality Index (PSQI).¹⁸ In another study, with 639 older adults from Taiwan, baseline social isolation (measured with a 4-item index: not being married, living alone, living in a rural area, not participating in any social groups) predicted worse PSQI scores 6 years later.²³ Two studies from the National Social Life, Health and Aging Project in the US, with 524²⁰ and 727¹⁹ subjects each, found that, while increasing social participation (volunteer work, religious services, meetings of organized groups) during 5 years was not associated with improved actigraphic measures of sleep quality or self-reported sleep characteristics,²⁰ having a positive marital relationship predicted better actigraphy-estimated sleep quality. Also, in a cohort of 1417 older adults in Singapore, older adults with a weak social network, assessed with a modification of the Lubben’s social network scale, showed an increased risk of restless sleep 2 and 6 years later, as measured with the item “my sleep was restless” of the Center for Epidemiological Studies Depression (CES-D) scale.²² Finally, in a cohort of 1951 British older adults, a greater quality of relationship with the closest person, measured with an adapted version of the Close Person’s Questionnaire, was related to better score on the PSQI up to15 years latter.²¹

Weak social networks may influence sleep through several biological pathways, such as the dysregulation of the hypothalamic-pituitary-adrenocortical (HPA) axis,^56,57 or the promotion of pro-inflammatory cytokine production.^58,59 Different studies have found a link between measures of social isolation, loneliness, and perceived social support in older adults, and hyperactivity of the hypothalamic–pituitary–adrenal (HPA) axis. For instance, in a study from the Medical Research Council National Survey of Health and Development with 1795 older adults, those who were widowed or who were living alone had higher nighttime cortisol and flatter diurnal slope than their counterparts.²¹ In another study with 78 Chinese older adults, those who spent more time and effort in developing and strengthening their social ties showed a greater rise in cortisol in the morning and a significantly steeper decline over the day, attesting to more effective activation and deactivation of the HPA axis.⁵⁶ Also, in a study with 92 African American older adults, a positive association between the frequency of weekly social activities and a steeper diurnal cortisol decline was observed.⁶⁰ Moreover, in the Longitudinal Study of Ageing, lower positive support was recently associated with higher hair cortisone levels in a sample of 2520 older adults.⁶¹ On the other hand, feelings of loneliness and a lack of social support have been associated with increased levels of inflammatory biomarkers like fibrinogen^62,63 or C reactive protein.^58,63 Also, loneliness has shown to heighten feelings of vulnerability and unconscious vigilance for social threat,⁶⁴ and increase total peripheral resistance and systolic blood pressure.⁶⁵ More specifically, we recently found that lower social support was associated with higher nocturnal systolic blood pressure, and lower systolic blood pressure dipping in older individuals.⁶⁶ Finally, a lack of social support may affect sleep hygiene practices, enhance chronic musculoskeletal pain,⁶⁴ or increase the adoption of unhealthy lifestyle-behaviors.^67–69

Limitations of our study include the use of self-reported sleep quality and sleep duration, which may be affected by recall and social desirability biases. However, the use of self-reported measures of sleep is habitual in large prospective studies due to its low-cost and simplicity. Although objective measures of sleep duration are more accurate, information related to personal perception of sleep quality can only be achieved with self-reported data. We also did not have information on sleep diaries. While we have not formally assessed the reliability of our instrument to measure social network, we have previously shown that this instrument was associated with hospital readmission and mortality among older patients with heart failure.³⁵

Conclusions

Poor social network is prospectively associated with short sleep and worse sleep quality. Given the large number of older adults with social isolation, and the high prevalence and health consequences of sleep problems, there is a need to develop and evaluate interventions (ie, providing company, emotional support) that target social isolation for improving this important health outcome.